AQR Capital Management, founded in 1998 by Cliff Asness and colleagues from Goldman Sachs Asset Management's quantitative research group, is the most academically flavoured of the large systematic managers. The firm publishes research, employs a small army of PhDs, and built its identity on factor investing: value, momentum, carry, defensive. The interview reflects that DNA, leaning on econometrics, model judgment and research taste more than on puzzle speed.
AQR interviews like the academy reviews a paper: they care how you handle models, not just whether you know them. The reported questions are about model selection, data handling and what you would do when a bigger model looks better, which is a judgment call dressed as a technical one.
- 1Recruiter and phone screens. Background and first technicals.
- 2Coding or modelling exercise. A practical assignment testing how you build and evaluate a model.
- 3Research-fit and behavioural conversations. Whether your interests match the group.
- 4Superday. A reported one-day, multi-round format with short timed exams and up to six interview rounds; PhD candidates present a job-market paper.
Stages 1 and 2: Screens and the exercise
The early technicals cover probability and statistics, econometrics, linear regression and hypothesis testing, and ML fundamentals. Reported phrasings include describing stochastic gradient descent and giving two ways to minimise a model's error, which is the closed-form-versus-iterative trade-off in disguise.
How to prepare
The first problem below was built from this firm's reported optimisation theme, and the others cover the econometrics layer the same rounds test:
Stage 3: The model-judgment rounds
The best-documented AQR question is a research-judgment case:
“How would you handle a 25-factor model differently from a 100-factor model, and what if the 100-factor model performed 25% better?”
What it testswhether better in-sample performance impresses you or worries you.
The strong answer treats the bigger model's outperformance as a claim to audit, not a result to celebrate: more factors mean more capacity to fit noise, so the burden of proof rises with the parameter count. Expect to discuss regularisation and factor selection, out-of-sample and walk-forward evaluation, and the multiple-testing arithmetic that says the best of many specifications looks good by construction. The multiple-testing section of our validation course is this question in course form.
A 2025 candidate report adds model selection, data handling and stress testing as recurring themes, and factor investing itself is table stakes: know what value, momentum, carry and defensive mean and why anyone believes they pay.
Stage 4: The superday
The reported format is a full day: short timed exams plus up to six rounds. For PhD candidates, a job-market-paper presentation anchors the day, and the defence of it is weighted heavily; the project-defence lesson is the drill for exactly that room.
Key tips for success
- Read some AQR research before interviewing. The firm publishes openly, the papers signal the house style, and referencing them well lands.
- Frame every model answer around validation: in-sample enthusiasm is the precise failure mode this firm screens against.
- PhDs: your paper is the interview. Run the twenty-question checklist against it before the superday.
Closing remarks
Within the three types of researcher process, AQR sits at the academic end of project-driven: econometrics, model judgment, and a paper defence at the end. Prepare like a referee rather than a contestant, and the factor-model question becomes the easiest kind there is: one you have already thought about.