Arrowstreet Capital is a systematic global equity manager in Boston, founded in 1999 by Bruce Clarke, John Campbell and Peter Rathjens and fully employee-owned since 2008. It runs long-only, alpha extension and long/short equity strategies for institutional clients, reports over $344bn in assets and more than 500 employees, and has a London entity. John Campbell is a Harvard economist, and the research culture is econometric first. Every campus role we found is in Boston, and the 10-week internship feeds full-time research hiring.
Arrowstreet screens research candidates with 3 separate online assessments, in maths, statistics and coding, and candidates report that their content leans heavily on portfolio optimisation and econometrics rather than speed arithmetic. Preparing only for traditional maths and statistics questions is not enough. No candidate has confirmed the vendor, the per-test lengths or the question counts.
- 1Online assessments. 3 tests: maths, statistics and coding.
- 2Recruiter call. About 30 minutes on your resume and why Arrowstreet.
- 3Technical interviews. 1 or 2 video rounds of 60 to 100 minutes on regression, probability and pandas, sometimes with live coding.
- 4Boston superday. A full day of around 8 short sessions, with a written test reported.
The 2027 programmes
| Programme | Who | Notes |
|---|---|---|
| Quantitative Researcher Intern, summer 2027 | Undergraduate or graduate students in finance, maths, economics or related, finishing within a year of the internship | Boston. Opened 13 July 2026, listed until 31 December 2026. Asks for probability, statistics, regression, time series, linear algebra, optimisation and portfolio theory, plus Python, R, Stata or MATLAB. New for 2027: experience with large language models and coding agents is a plus. $3,500 to $5,000 per week |
| Quantitative Developer Intern, summer 2027 | CS or engineering plus data science, applied maths or economics | Boston. Software design, data structures, numerical algorithms, regression and time series. Same pay |
| Quantitative Researcher, full time | Enrolled in or graduated from an undergraduate or graduate programme. Recent graduates are encouraged to apply | $155,000 to $260,000 base range posted |
| Quantitative Developer, full time | 1 to 3 years of experience, Python with pandas and numpy | Same range |
| Technology Rotational Program and Investment Services Rotational Program | Graduates | 18 months and 2 years |
A PhD is not required for research. A separate PhD-titled researcher listing existed in 2025, alongside the undergraduate route rather than instead of it. Candidate reports cluster from September to December.
Stage 1: Online assessments
| Item | What candidates report |
|---|---|
| Number of tests | 3: maths, statistics and coding, in candidate reports from 2023 to December 2025 |
| Order | Not reported consistently |
| Vendor | Unconfirmed. Possibly a HackerRank-style platform, but no candidate has said so. HackerRank CodePair is used at the next stage |
| Length | A 3-hour combined sitting, in 1 intern report. A secondhand account of the 2027 cycle describes about 13 questions at roughly 4 minutes each |
| Answer format | Typed numeric answers with explicit rounding, such as "rounded down to the nearest integer" or "in %", plus single-select and multi-select multiple choice |
| Coding difficulty | LeetCode easy to medium for research. Software engineering intern tests are far harder, 3 LeetCode hards, and the quantitative developer test is harder than at comparable firms |
| Proctoring, navigation, calculator, pass bar, invitation, window, retakes | No reports |
Two versions may be circulating. Candidate reports to December 2025 describe 3 separate tests, and the secondhand 2027-cycle account describes 1 combined screen of numeric and multiple-choice questions. Either the format changed, or the account covers 1 of the 3 tests. Expect 3 invitations and read each email. If you have sat them recently, tell us what you saw.
What is on each test
- Maths. Calculus, linear algebra and optimisation. Portfolio optimisation and allocation is the confirmed theme: tangency portfolio and factor-model Sharpe ratios, the beta of an asset to an equal-weighted portfolio, the volume of a constrained cone, the KKT system of a trade-scheduling problem, and a linear programme.
- Statistics. Regression and time series, with portfolio calculations mixed in. Reported items: the distribution of a sum of normals, a likelihood-ratio test between 2 decks of cards, and multiple choice on dummy-variable specification, adding quadratic terms, calendar-month train and test splits, seasonal time series models, and what a t-statistic of 4 implies.
- Coding. LeetCode easy to medium in Python.
How to prepare
- Portfolio maths by hand. Mean-variance and tangency weights, the Sharpe ratio of a combined portfolio, beta to a portfolio, Lagrangians and KKT conditions, small linear programmes. The optimisation puzzles lesson is the starting point.
- Econometric judgement. OLS assumptions and their violations, multiple regression with dummies and nonlinear terms, stationarity and autocorrelation, seasonal models, and why k-fold fails on time-ordered data.
- Distributions. Sums of normals and likelihood ratios, from the multivariate normal lesson and hypothesis testing.
- Coding. LeetCode easy and medium in Python with pandas, from the coding practice catalog.
- Typed answers. Practise computing to a stated rounding rule, because the answer boxes ask for it.
Stage 2: Recruiter call
About 30 minutes by phone: your resume, why Arrowstreet, and how you heard about the firm. Candidates describe quant researcher processes moving in about a week, so the stages can come quickly once the tests are in.
Stage 3: Technical interviews
Video rounds with a senior researcher, sometimes a researcher and a developer together. Reported lengths are 60 minutes, 80 minutes with 2 interviewers plus 20 minutes of HR, and 3 back-to-back sessions of 2 technical interviews of 45 minutes plus 15 minutes of behavioural questions. Linear regression comes up for several candidates.
What they cover
- Regression. The assumptions of OLS, what happens when they are not satisfied and how you handle it. Candidates say regression questions come up a lot.
- Finance. CAPM, factor models and some statistics.
- Probability and brainteasers. Green Book probability, several brainteasers, and a binary search problem at LeetCode medium.
- Python and pandas. A pandas data interview, and on CodePair, Python slicing and aliasing such as the output of a negative-step slice, list slice copies against numpy views, static methods against class methods and inheritance, pandas merges with duplicate keys, groupby aggregate against transform, rolling windows with a minimum period, and resampling. Then 2 LeetCode easy or medium problems.
- Your resume. The machine learning models on it, in detail.
Full-time quantitative developers can face a 90-minute CoderPad round, all in Python, and it is demanding. A reported question is the maximum depth of an n-ary tree without recursion.
How to prepare
State the OLS assumptions and the fix for each violation without notes. Work the quant Python lessons until slicing, views and pandas groupby semantics are second nature, and the probability questions for the Green Book round. Know CAPM and factor models well enough to derive a beta.
Stage 4: Boston superday
In person, a full day at the office, with lunch. Expect up to 8 sessions of 30 minutes, 1-on-1 or 2-on-1. Several candidates describe a written or coding component in the final. An experienced developer describes a written programming test followed by roughly 10 people interviewing over a full day. A secondhand summary of the written exam lists the mean-variance tangency portfolio, risk parity and equal risk contribution, regressing Y on X against X on Y, effective degrees of freedom under autocorrelated residuals, log utility and the certainty equivalent, and one asset-pricing item. No candidate describes a research presentation round, and no games, cases or take-homes are reported.
How to prepare
Assume the day opens with a written exam done by hand and that the topics are the same portfolio and econometrics families as the tests. Then expect 8 short conversations that revisit regression, statistics, finance, coding and fit. Pace yourself and treat each session as a fresh start.
Key tips for success
- Prepare portfolio optimisation and regression specification, not mental arithmetic. Nothing reported resembles a speed test.
- Expect 3 test invitations and give each its own sitting with the rounding rules in front of you.
- Learn pandas semantics precisely. The CodePair questions are about what the library actually does.
- Keep your resume's models explainable to the equation.
- Move fast once the tests are in. Candidates describe researcher processes taking about a week.
Closing remarks
Arrowstreet's process is an econometrics exam in 3 parts, a regression conversation, and a Boston day that repeats both on paper and in person. The candidates who struggle are the ones who prepared for a trading firm's screen. The ones who do well can set up a tangency portfolio, diagnose a regression and write pandas without thinking about it.
Work through everything we have for Arrowstreet, and read our quant researcher interview process guide for how an asset manager's process compares with a market maker's.