Akuna Capital · Phone screen → Onsite
In processAkuna Capital QR Regression Rounds: 11 Experiences (Part 3 of 6)
Akuna Capital quant researcher interviews from 2020: 11 reports on root finding, regression, shuffles and simulation coding.
At a glance
- Process (2020): one or two phones, then a final round that mixed proofs, coding and ML or regression discussion (H32–H42).
- Asked in these reports: coding, probability and statistics/regression in 8 of 11 each; ML in 5; examples include root finding (H35), shuffle proofs (H34, H38) and an airplane-seat simulation (H32).
- Difficulty: one candidate misread a mean-versus-median question across two phones (H41).
Coverage: 11 Akuna Capital quantitative research reports from March to December 2020 (QR intern, QR, Junior QR, QR-ML; Chicago where stated). All are more than three years old, so they sit in History below.
In this series
- Part 1: 2022–2025 screens, coding and the newest reports.
- Part 2: 2021–2022: coding, simulation and probability.
- Part 3 (current): 2020: root finding, regression and shuffles.
- Part 4: 2019–2020: ML and final rounds.
- Part 5: 2018–2019: probability and SVMs.
- Part 6: 2017–2018: JQR, ML and projects.
Newer reports: Akuna quant technical interviews, 2025–2026.
What came up
Regression and statistics came up as often as coding and probability in 2020, with several rounds asking candidates to prove a result and then simulate it. Every question stays in its entry; there are no current tables.
Common questions
What statistics questions did Akuna ask quant researchers in 2020?
Regression came up in 8 of 11 reports, from a long regression discussion (H35) to AR and MA models (H39) and PCA and covariance (H40).
Did Akuna ask simulation coding?
Yes in several 2020 reports: an airplane-seat simulation (H32), final-round simulations after shuffle proofs (H34) and a simulation-only coding task (H37).
History (before 2024)
H32: Loss functions, a sphere and an airplane simulation
Report date: 2020-12-09 · Role: Quantitative Researcher
HR arranged a phone after a long wait. Resume discussion and two mathematical questions came before a simulation written on CodePair.
Technical phone
- Given n real numbers, choose x to minimize the sum of squared distances and justify it. Then replace squares with absolute distances and explain the new minimizer.
- Choose a point uniformly on a three-dimensional sphere of radius r. What is the variance of its x-coordinate?
- Model the airplane-seating problem: the first passenger takes a random seat; later passengers take their own seat when available and otherwise a random free seat. Find the probability that the last passenger gets their own seat, then write a simulation on CodePair.
H33: Two mathematics questions and a sixth-degree equation
Report date: 2020-12-04 · Interview date: 2020-12-03, inferred from the next-day report · Role: Junior Quantitative Researcher
The interview was planned for 45 minutes but lasted about an hour while the candidate worked through the questions.
Technical phone
- Which is larger: e^π or π^e?
- For a and b uniform on (0,1), find the probability that x² + ax + b = 0 has a solution. The account does not explicitly state independence or the real-root convention.
- Write code to solve a sixth-degree polynomial numerically. The candidate did not retain its coefficients or a required numerical method.
H34: Shuffle proofs, random generators and final-round simulations
Report date: 2020-11-20 · Interview period: mid-October and mid-November 2020 · Role: Junior Quantitative Researcher
The phone was in mid-October and the virtual onsite in mid-November 2020. After HR, the final schedule was expected to contain five technical or manager conversations, but the interview ended after four.
Technical phone
- Shuffle an array, improve the approach, and prove why Fisher–Yates works.
- Generate rand10 from rand7. Find the expected number of draws, then derive the geometric-distribution expectation used in the calculation.
Final interview 1
Final interview 2
- Given probabilities for links between pairs of nodes, simulate the probability that the resulting graph is connected. Use hypothesis testing to assess the result.
Final interview 3
- Eigenvalues came up in a later discussion.
- PCA was also discussed.
- Another topic was time-series analysis.
- The interviewer asked about Python decorators.
- Tree models were on the list too.
- Correlation was another reported topic.
Final interview 4
- Differentiate x^x.
H35: Root finding and a long regression discussion
Report date: 2020-10-22 · Role: Junior Quantitative Researcher, ML · Location: Chicago
The Chicago final covered coding, regression, internship work, geometry and HR. The first interviewer was 15 minutes late. The regression discussion lasted 45 minutes, and a rejection came after the candidate followed up about a week and a half later.
Final interview 1
- Write code to solve x^N = a. This account did not specify gradient descent as the required method.
Final interview 2
- Spend about 45 minutes discussing the resume and linear regression, including when to choose L1 or L2 regularization.
Final interview 3
- The interviewers went closely through the candidate's internship work and challenged its technical choices.
Final interview 4
- An elementary geometry question involved a square, circle and lines, asking for a small region's area. The candidate no longer remembered the diagram or measurements.
H36: Research followed by a uniform-sum proof
Report date: 2020-10-16 · Role: Quantitative Research Intern
An alumnus referred the candidate for a summer internship. Detailed research discussion used much of the phone before the final probability-proof question.
Technical phone
- The interviewer explored the candidate's previous research in detail.
- For N independent U[0,1] variables, find and prove the probability that their sum exceeds 1.
H37: Noisy predictors and a simulation-only coding task
Report date: 2020-10-01 · Role: Quantitative Research Intern
The first phone moved from regression discussion into a coding simulation.
Technical phone
- In a regression of Y on X, what happens to the fitted slope if white noise is added to X? The candidate first said it stayed unchanged, then corrected themselves: the covariance numerator stays the same while the variance denominator grows.
- Simulate the airplane-seating problem and estimate the chance that the last passenger gets their own seat. This interviewer did not ask for an analytical solution.
H38: A shuffle and an array median
Report date: 2020-09-30 · Role: Junior Quantitative Researcher, ML
HR arranged a one-hour phone directly. Resume and ML discussion were followed by two coding tasks.
Technical phone
- The resume discussion included general ML knowledge, without retained individual questions.
- Randomly shuffle an array.
- Find the median of an array.
H39: AR, MA and two coin questions after a later update
Report date: 2020-08-09 · Interview window: after 2020-08-09 and by the 2020-09-29 update · Role: Quantitative Research Intern · Location: Shanghai
The Shanghai research-internship candidate initially had a phone coming up. A September 29 update confirmed that the English-language interview had happened and ended in rejection.
Technical phone
- What is an autoregressive, or AR, model?
- What is a moving-average, or MA, model?
- There were two coin-based probability questions. The candidate did not recall the setups, but said the conversation emphasized explaining the reasoning rather than lengthy calculation.
H40: Neural networks, PCA and covariance
Report date: 2020-08-05 · Role: Junior Quantitative Researcher
Introductions and previous research or work projects led into basic ML questions. The candidate thought they had failed.
Technical phone
- How does a feed-forward neural network work?
- Explain the principle behind PCA and how to carry it out.
- How are covariance and correlation related?
H41: Two phones and a mean-versus-median misunderstanding
Report date: 2020-03-23 · Role: Quantitative Research Intern
There were two half-hour phones in English: mathematics first, then online coding. The candidate reported eventual rejection.
First phone
- The first phone covered linear algebra and probability/statistics.
- The candidate later recalled the law of large numbers.
- They also recalled the central limit theorem.
Second phone
- The second phone included basic list-comprehension questions, without the expressions.
- Build a class that accepts data and returns the mode and mean. The candidate initially mistook mean for median.
H42: Dated phones and an onsite spanning math, code and ML
Report date: 2020-03-11 · Interview dates: 2020-02-05, 2020-02-12 and 2020-03-04 · Role: Quantitative Researcher, ML · Location: Boston
Phones on February 5 and February 12 were followed by a Boston onsite on March 4, 2020. That day included a roughly 20-minute project presentation and remote Chicago interviewers. A planned sixth coding round was cancelled after the fifth completed round; rejection arrived March 9.
First phone: February 5
- For Gaussian X, find P(X² + 2X < 3). The full distribution parameters were not retained.
- How can you determine whether a matrix has full rank?
- Explain a determinant as volume in higher dimensions.
- Why is the constant e important?
- A hash function produces values in 0–3. How would you enlarge its range gradually? The candidate recalled an explanation involving a change of modulus, but not enough detail to settle the intended construction.
Second phone: February 12
- Remove duplicates from a Python list of numbers.
- Use a list comprehension to return powers of 2 in an integer interval [a,b].
- Implement a straightforward class; its actual functionality was not recorded.
Onsite: project presentation
- Give a 20-minute presentation on a resume project, with questions from the staff throughout.
Onsite: mathematics
- Draw one of five equally likely colors with replacement. Find the expected number of draws or days needed to see every color, then derive the geometric expectation used.
- A toilet-paper roll has initial radius R and unwinds at constant speed v. Find its radius r as a function of time. The interviewer supplied a differential equation and asked the candidate to solve it.
Onsite: class coding
- Complete a stock-trading-system class in code. The candidate called the round system design, then clarified that it was coding and only similar to another suggested trading-system problem.
Onsite: ML discussion
- How would you handle correlated features in linear regression? The candidate's PCA answer did not satisfy the interviewer.
- Compare XGBoost, AdaBoost and gradient boosting.
- How would you turn classification into regression? The candidate suggested label encoding and was asked for alternatives.
- Compare Newton's method with gradient descent beyond their convergence behavior.
Keep practicing
- From these reports: shuffle and reset array (H34), Use gradient descent to solve x^n = A (H34), choose L1 or L2 regularization (H35), Find the median of an array (H38), stream max mean mode (H41), remove duplicates preserving order (H42).
- Akuna Capital question banks: algorithms (sign-in needed).
- Everything Akuna Capital in one place: the Akuna Capital interview page.
- More firsthand reports: the Experience feed.
In this series
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.