Company interview guide

AQR Capital Management Machine Learning Interview Guide (2026)

For ML, AI and research roles: a statistics-first systematic fund where regression comes up in nearly every round, with coding screens and no trading games.

AQR Capital Management's machine learning interview loop has 7 reported stages, from application & recruiter or fit call to hiring-manager & fit conversations; typical length: Under a month for some campus cycles, usually 1-2 months with gaps.

Roles covered: Research Scientist, Data Scientist, ML Engineer and Applied Scientist. The question bank holds 7 questions reported from AQR Capital Management interviews, 4 of them free to solve in your browser.

Try a free question

Interview process length
Under a month for some campus cycles, usually 1-2 months with gaps
Stages
7
Reports linked to bank questions
7
Bank report dates
Jun 2017–Jul 2020
Questions in bank
7 (4 free)
Last reviewed
October 2026

How the AQR Capital Management interview loop works

Apply through the firm's careers site (summer internship applications open at the end of June and interviews run through September), then a recruiter or fit call. Most candidates next take an online test: engineering-track candidates a 75-90 minute, three-problem coding test, research-track candidates either the same kind of coding test or a recruiter-sent math screen of about eight multiple-choice questions. Research-track candidates then have one to three technical phone rounds of about an hour mixing statistical inference, regression and coding, often with a deep dive on past projects, followed by a superday of four to six interviews of about 45 minutes, sometimes with written exams. Engineering-track candidates do a week-long design take-home and an on-machine onsite. A hiring-manager or fit conversation closes most loops. The firm runs no market-making, betting or mental-math games on any track. The outline rests on about ten first-hand candidate reviews, none for an ML-titled role.

The AQR Capital Management interview process, stage by stage

01Application & recruiter or fit call

Apply on the firm's careers site or through campus placement; then about 30 min with HR or a team member on your resume, motivation and fit

Most loops start with a short call on your background and why you want a systematic, research-driven fund. Internship applications for the following summer open at the end of June and close in mid-August, with interviews through September; some international campuses apply through the university placement office. Compiled accounts note that work authorization and visa timelines get close attention at this stage.

What it tests

  • Why a systematic, academically minded fund and why this team
  • How your resume maps to research, implementation or engineering work
  • Logistics: track, location, timing and work authorization

How to prepare

  • Ask which group and track you are interviewing for: research and portfolio management, portfolio implementation, risk or engineering; the track decides the next test.
  • Be ready to explain one project in two minutes with the numbers that mattered.
  • Apply only through the firm's own site or campus office; the firm warns that impostors contact people over messaging apps.

A resume, motivation or fit call opens 3 research-track candidate reports (2021-2022) and 1 other-role report. The firm's careers and internship pages describe applying and the interview window.

02Online test: coding test or math screen

Coding: 75-90 min, 3 problems from a rotating bank on HackerRank or CodeSignal.

Math screen (some research candidates): a recruiter-sent take-home with an algorithms document and about 8 GRE-level multiple-choice questions plus 2 bonus, no calculator

The coding test long used a fixed trio: raise p to the power q faster than linear time without overflow, sort a million integers drawn from 1 to 10, and split a string at digit boundaries, sort each run and join. Newer sittings draw from a wider bank: counting islands or friend circles with union-find, range queries for the largest d[i] - d[j] with i before j (naive loops time out), bounded knapsack, jump-game DP and palindrome insertions. The math screen covers geometry, counting, probability and algebra with no calculus; careful reading is the main hurdle. Some research candidates also describe a skills test plus an IQ or personality test.

What it tests

  • Picking the complexity that fits the constraints before coding (sparse tables, counting sort, fast exponentiation)
  • Union-find and grid traversal
  • DP with a capacity or jump limit
  • Short probability, counting and algebra questions read precisely

How to prepare

  • Budget 25-30 minutes per coding problem and estimate the operation count first; the range-query problem is built to time out a nested loop.
  • Watch integer overflow in the exponentiation and counting problems.
  • On the math screen, reread each condition before answering; the math is easy, the wording is not.

Format and problem bank from compiled accounts; 1 research-track report (2025) had an online test and 2 (2021-2022) a skills test. The firm's campus internship page lists an online test first.

03Technical phone rounds: statistics, regression & coding

1-3 rounds of about 60 min with an associate, researcher or MD; often 2 back to back; research track mixes inference, regression, live pandas or statsmodels and coding, engineering track covers CS fundamentals

Regression is asked in almost every research round: state the OLS assumptions, derive the estimator, say what breaks when an assumption fails, explain omitted-variable bias and write a covariance function. Other recurring prompts: derive a maximum-likelihood estimate, remove outliers and recompute the mean and standard deviation, tell whether a return series has been time-reversed from its skew (indices skew negative, single names positive), and walk through how a factor in your own work was built and why. The most recent research report (2025) instead had three phone rounds of medium-to-hard algorithm problems with runtime follow-ups. Engineering-track phones cover OOP and SOLID, SQL indexes and complexity, often without live coding.

What it tests

  • OLS: assumptions, derivation of beta, diagnostics and fixes
  • Maximum-likelihood derivations for a given distribution
  • Outliers, skewness and covariance computed by hand or in pandas
  • Algorithm problems with an exact runtime analysis
  • Defending how and why you built a feature or factor

How to prepare

  • Derive OLS from the normal equations and list each assumption with what failing it does to bias and to standard errors.
  • Be able to fit and read a regression in statsmodels and do a groupby or rolling window in pandas without looking anything up.
  • Prepare a five-minute walk-through of one project that covers data, method choice, validation and what you would optimise next.

Free reported questions for this stage

Technical phone rounds in 5 research-track candidate reports (2015-2025): 3 on statistics and regression, 2 mostly coding. Round count and topics also follow compiled accounts of the quant research loop.

04Design take-home: PnL calculator (engineering track)

About one week; build a profit-and-loss calculator from two message files

The engineering track's signature step: parse a trades file and a prices file and compute profit and loss for long and short positions. Graders weigh object-oriented design, structure, code style and comments, not just correct numbers, and edge cases such as partial fills and message ordering matter.

What it tests

  • Clean class design for positions, trades and the calculator
  • Long and short handling, partial fills and out-of-order messages
  • Readable, tested, documented code

How to prepare

  • Write down your assumptions about the file formats and edge cases in a short README.
  • Separate parsing, state and reporting so each can be tested alone.
  • Add unit tests for short positions and position flips; they are the usual bugs.

From compiled accounts of the firm's software loop only; no research-track or ML-role candidate report mentions a take-home.

05On-machine onsite (engineering track)

A 2-3 hour on-machine project reviewed by an engineer, then 3-4 rounds of about 45 min on algorithms and behavior

The morning project is a task scheduler: tasks with dependencies, durations and allowed time windows, solved with a topological sort plus a heap and then walked through with an engineer. Later rounds cover binary search over a continuous range to invert a monotone function, tree paths that sum to a target and may bend through a node, a fixed-size container with constant-time insert and delete, and standard graph and DP problems.

What it tests

  • Topological sort combined with a priority queue under time constraints
  • Binary search on real numbers with a precision stopping rule
  • Constant-time data structures such as an LRU-style cache
  • Tree and DP problems explained with complexity

How to prepare

  • Practise a dependency scheduler end to end on your own machine, with tests, inside two hours.
  • For continuous binary search, state the tolerance and the loop condition before you code.
  • Keep the project code tidy; an engineer reviews it with you afterwards.

Free reported questions for this stage

From compiled accounts of the firm's software loop only; the two practice problems come from older software-role tags, and no research-track report describes this onsite.

06Superday: statistics, ML, finance & projects (research track)

4-6 one-on-one interviews of about 45 min with team members, in person or virtual; some groups add written finance, math and programming exams or a deliberately stressful interview

Each interviewer takes a topic: statistical inference, ML fundamentals, finance, your projects or behavior. Reported ML questions cover lasso and ridge (and how to pick lambda), PCA, random forests and boosting, and overfitting; statistics questions cover fixing violated OLS assumptions, covariance and why daily and monthly correlations differ. Finance covers CAPM, Sharpe ratio, value at risk (including for an option), credit spreads and bond duration. Coding is light here: sorting complexity, two pointers, sliding windows or a covariance function on a whiteboard.

What it tests

  • Regularization: lasso vs ridge, choosing lambda, bias and variance
  • PCA, tree ensembles and overfitting on noisy data
  • Repairing OLS when assumptions fail; covariance and correlation across horizons
  • Finance basics: CAPM, Sharpe, VaR, duration
  • Depth on your own research

How to prepare

  • Prepare a one-line intuition and a derivation for lasso, ridge and PCA; interviewers ask for both.
  • Know why correlation measured on daily returns differs from monthly (noise, asynchronous closes, autocorrelation).
  • If one interviewer turns hostile, stay calm and keep reasoning aloud; candidates describe it as a temperament check.

A final round or superday in 5 research-track candidate reports (2021-2022), four to six interviews where stated; written exams and the stress round come only from compiled accounts.

07Hiring-manager & fit conversations

A final conversation with the hiring manager or a senior member, plus behavioral questions spread through the superday

Most loops close with the hiring manager or a senior researcher. Questions are mostly about motivation, why a systematic fund, how you work with others and how you handle being wrong; candidates describe the tone as conversational. Behavioral questions also appear inside superday rounds.

What it tests

  • Why the firm and why systematic investing
  • Collaboration across research, implementation and engineering
  • Intellectual honesty about failed ideas

How to prepare

  • Read one or two of the firm's public research papers in your area and have a view on them.
  • Prepare a story about an idea that did not survive testing and what you learned.
  • Have questions ready about how research moves into live portfolios.

A hiring-manager or fit conversation in 3 research-track candidate reports (2021-2025) and 1 other-role report, where the main question was why the firm.

Questions and themes reported most often

  • Derive the OLS estimator, list its assumptions and say what breaks when each fails (3 reports, technical phone rounds: statistics, regression & coding)
  • Algorithm problems with exact runtime analysis: sorting, two pointers, sliding windows, primes (3 reports, technical phone rounds: statistics, regression & coding)
  • Lasso, ridge, PCA and random forests: intuition, tuning and when each helps (1 reports, superday: statistics, ml, finance & projects (research track))
  • Derive a maximum-likelihood estimate for a given distribution (1 reports, technical phone rounds: statistics, regression & coding)
  • Covariance and correlation: write the function, prove a covariance matrix is PSD, compare daily and monthly correlation (1 reports, superday: statistics, ml, finance & projects (research track))
  • Value at risk for a position, including an option (1 reports, superday: statistics, ml, finance & projects (research track))
  • How a factor was constructed and why that method was chosen (1 reports, technical phone rounds: statistics, regression & coding)
  • Omitted-variable bias and collinear regressors (technical phone rounds: statistics, regression & coding)

What AQR Capital Management asks about

  • Statistics & regression30%
  • Coding & algorithms22%
  • ML fundamentals & validation18%
  • Probability & math screen10%
  • Finance & factor reasoning10%
  • Projects & behavioral10%

AQR Capital Management interview prep checklist

  • Derived OLS from the normal equations and explained each assumption, its failure and its fix
  • Derived a maximum-likelihood estimate and explained omitted-variable bias with an example
  • Fit and interpreted a regression in statsmodels and did rolling statistics in pandas live
  • Explained lasso vs ridge, choosing lambda, PCA and random forests with a finance example each
  • Set up purged, time-ordered validation and named three backtest leakage traps
  • Solved a three-problem timed coding mock in 75 minutes, including a range-query problem
  • Can explain CAPM, Sharpe ratio, value at risk and how a factor is constructed
  • Prepared a project walk-through: data, method choice, validation and what failed

Tips

  • Settle your track in the first call: research groups run a statistics-heavy loop, engineering groups a software loop with a take-home.
  • Make regression routine: derive OLS, list assumptions, and fix each violation aloud, since it comes up in almost every research round.
  • Practise live pandas and statsmodels: groupby, rolling windows, fitting and reading a regression without documentation.
  • Do not skip algorithms; the most recent research report was mostly medium-to-hard coding with runtime questions.
  • Prepare intuition plus derivation for lasso, ridge, PCA and tree ensembles, and say when each would overfit a return series.
  • Know enough finance to talk fluently: CAPM, Sharpe ratio, value at risk, duration and how a value or momentum factor is built.
  • Frame ML experience around out-of-sample validation, leakage control and small, fragile effects rather than model size.
  • Skip trader-style prep such as market-making games and mental-math drills; candidates report none on any track.
  • Plan for a one-to-two-month process with gaps, and raise work-authorization questions early.

See the rest of the AQR Capital Management loop

The later stages, the questions reported for them and how to prepare are in Pro.

Unlock with Pro

Practice the AQR Capital Management loop

Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.

AQR Capital Management interview FAQ

How many interview rounds does AQR Capital Management have for ML roles?

Candidates report 7 stages: Application & recruiter or fit call, Online test: coding test or math screen, Technical phone rounds: statistics, regression & coding, Design take-home: PnL calculator (engineering track), On-machine onsite (engineering track), Superday: statistics, ML, finance & projects (research track) and Hiring-manager & fit conversations. Apply through the firm's careers site (summer internship applications open at the end of June and interviews run through September), then a recruiter or fit call. Most candidates next take an online test: engineering-track candidates a 75-90 minute, three-problem coding test, research-track candidates either the same kind of coding test or a recruiter-sent math screen of about eight multiple-choice questions. Research-track candidates then have one to three technical phone rounds of about an hour mixing statistical inference, regression and coding, often with a deep dive on past projects, followed by a superday of four to six interviews of about 45 minutes, sometimes with written exams. Engineering-track candidates do a week-long design take-home and an on-machine onsite. A hiring-manager or fit conversation closes most loops. The firm runs no market-making, betting or mental-math games on any track. The outline rests on about ten first-hand candidate reviews, none for an ML-titled role.

How long does the AQR Capital Management ML interview process take?

Reported timelines run Under a month for some campus cycles, usually 1-2 months with gaps.

What does AQR Capital Management ask in ML interviews?

By reported question mix: Statistics & regression 30%; Coding & algorithms 22%; ML fundamentals & validation 18%; Probability & math screen 10%; Finance & factor reasoning 10%; Projects & behavioral 10%. The most repeated themes: Derive the OLS estimator, list its assumptions and say what breaks when each fails; Algorithm problems with exact runtime analysis: sorting, two pointers, sliding windows, primes; Lasso, ridge, PCA and random forests: intuition, tuning and when each helps.

How should I prepare for the AQR Capital Management ML interview?

Settle your track in the first call: research groups run a statistics-heavy loop, engineering groups a software loop with a take-home. Make regression routine: derive OLS, list assumptions, and fix each violation aloud, since it comes up in almost every research round. Practise live pandas and statsmodels: groupby, rolling windows, fitting and reading a regression without documentation.

Are there free AQR Capital Management interview questions?

Yes. LeetML has 7 questions reported from AQR Capital Management interviews, and 4 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources