RepostedAkuna Capital logo

Akuna Capital · Phone screen → Onsite

In process

Akuna Capital Quant Oral Rounds: 11 Experiences (Part 3 of 3)

Quant (general) · UnknownReported Oct 4, 2021
Shared byjanebeingjane

Akuna Capital quant oral technical interviews from February 2020 to October 2021: 11 reports across QR, trading, quant dev and market risk roles.

At a glance

  • Process (2020–2021): live phones, interactive video and onsite rounds; one Boston loop had two phones and an onsite (H23–H33).
  • Asked in these reports: probability in 7 of 11, statistics and ML in 5, linear algebra and Python concepts in 4 each; examples include absorbing walks and random hats (H24), hash collisions and Python memory (H27) and differentiating x^x (H30).
  • Difficulty: one candidate's uniform-sum answer was not accepted without a proof (H31).

Coverage: 11 Akuna Capital reports from February 2020 to October 2021: QR, QR intern, quantitative trading intern, junior quant developer and a market risk quant; Shanghai and Boston where stated. All are more than three years old, so they sit in History below.

In this series

  • Part 1: the newest oral technical reports.
  • Part 2: October 2021 to August 2022.
  • Part 3 (current): the oldest oral technical reports.

What came up

A wider mix of roles than the other parts, so the questions range from regression derivations to Python internals, alongside the usual probability. Every question stays in its entry.

Common questions

What does Akuna ask quant developers in oral rounds?

In this part, Python and computing concepts such as hash collisions and memory layout (H27), next to probability.

Does Akuna expect proofs in quant interviews?

One candidate's correct uniform-sum answer was not enough without a proof (H31).

History (before 2024)

H23: A structured matrix and a uniform-sum probability

Report date: October 4, 2021 · Role: Quantitative Trading Intern 2022 · Stage: First phone · Result: Passed this round

First two questions

  1. An n×n matrix has n on every diagonal entry and 1 elsewhere. Find every eigenvalue and eigenvector. The candidate gave eigenvalue 2n−1 along the all-ones vector, and eigenvalue n−1 on the subspace perpendicular to it.

  2. For n Uniform[0,1] random variables, find the probability that their sum is less than 1. The author gave 1/n!. Independence is needed for that value; the report’s short wording does not state it explicitly.


H24: A simplex, an absorbing walk, and random hats

Report date: October 1, 2021 · Role: QR Intern · Location: Shanghai · Stage: Phone

A roughly 45-minute screen

  1. For independent Uniform[0,1] variables, find P(X₁+…+Xₙ<1). Start with n=2 and n=3, then extend the reasoning to general n.

  2. A one-dimensional asymmetric random walk starts at 1 and has an absorbing boundary only at 0. Find the expected number of steps until absorption. The candidate felt the interviewer wanted a rigorous proof rather than intuition, possibly involving martingales. The direction probabilities and whether the expectation is conditioned on eventual absorption were not recorded.

  3. N people take hats at random. If X people get their own hats, find E[X] and Var(X). The candidate pointed to linearity of expectation; the requested variable counts correct matches.


H25: Even heads and an increasing run

Report date: September 27, 2021 · Role: QR Intern · Stage: Phone · Result: Rejected

A 40-minute screen

  1. After N fair coin tosses, what is the probability of an even number of heads?

  2. Draw successive values from Uniform[0,1] and stop at the first decrease. Find the expected length of the increasing sequence.

  3. The candidate initially misunderstood the second question. Afterward, they used symmetry to say that n draws are increasing with probability 1/n!, and wrote an answer of e. Counting the final decreasing draw gives e; counting only the increasing prefix gives e−1. The retained wording does not settle that counting boundary.


H26: Regression derivations in a Market Risk quant interview

Report date: August 13, 2021 · Role: Market Risk quant · Stage: Interactive video interview

Statistics and ML discussion

  1. Which Python packages have you used?

  2. Derive linear regression and logistic regression. Explain supervised learning, Lasso, and Ridge.

  3. The author also mentioned probability questions but did not identify them.


H27: Hash collisions and Python memory layout

Interview date: April 7, 2021 · Role: Junior Quant Developer · Stage: Phone · Result: No further response reported

Follow-ups and concepts

  1. After a duplicate-removal coding task, the interviewer asked how a Python set handles hash collisions.

  2. How are the contents of a NumPy array and a Python list stored in memory? The candidate said they were unfamiliar with this and skipped it.

  3. The author later updated the post to say they had heard nothing more.


H28: Mean versus median, a sphere, and airplane seats

Report date: December 9, 2020 · Role: QR · Stage: First phone

Mathematics and a probability follow-up

  1. Given n real numbers, choose x to minimize the sum of squared distances and explain why the mean works. Replace squared distance with absolute distance and explain the resulting choice.

  2. A point is uniformly distributed on the surface of a three-dimensional sphere of radius r. Find the variance of its X coordinate. The source specifies the surface rather than a solid ball.

  3. The first airplane passenger chooses a seat randomly. Each later passenger takes their assigned seat if available, otherwise a random available seat. What is the probability that the final passenger gets their own seat? The interviewer asked for an analytic value and then a simulation giving a numerical estimate. The author confirmed that CodePair was used.


H29: Comparing two powers and a random quadratic

Interview date: December 3, 2020 · Report date: December 4, 2020 · Role: Junior Quant Researcher · Stage: Phone

Mathematics portion

  1. Which is larger: e^π or π^e?

  2. What is the probability that x²+ax+b=0 has a solution when a and b are Uniform(0,1) variables? The post does not explicitly specify independence or say whether “solution” means a real solution.

  3. The call had been planned for 45 minutes but ran to about an hour because the candidate struggled with the questions. A later numerical coding task concerned a sixth-degree polynomial.


H30: Differentiating x^x in a manager conversation

Interview periods: Mid-October phone and mid-November 2020 onsite · Report date: November 20, 2020 · Role: Junior Quant Researcher, full-time · Stage here: Onsite

Later onsite rounds

  1. In the fourth technical or manager round, the interviewer asked the candidate to differentiate x^x.

  2. An earlier short-answer round touched on eigenvalues, PCA, time series, Python decorators, tree models, and correlation. The author retained those topic names rather than specific prompts.

  3. The candidate had been told to expect another round, but the day ended after this conversation. No question from the canceled round was reported.


H31: A uniform-sum answer was not enough without a proof

Interview period: That week in October 2020 · Report date: October 16, 2020 · Role: Quant Researcher; candidate seeking a summer internship · Stage: Phone

After a long research discussion

  1. Given N IID standard uniform variables, find the probability that their sum exceeds 1 and give a mathematical proof.

  2. The candidate could obtain an answer quickly but got stuck when asked for a proof. They tried conditioning arguments and were unsure how the interviewer judged the explanation.

  3. Research discussion used much of the call, leaving time for this one technical question.


H32: Neural networks, PCA, covariance, and correlation

Report date: August 5, 2020 · Role: Junior Quant Researcher · Stage: Phone

ML fundamentals

  1. How does a feedforward neural network work?

  2. What is the principle behind PCA, and how do you perform it?

  3. What is the relationship between covariance and correlation?

  4. These followed a short introduction to the candidate’s research and work projects. The candidate felt their ML fundamentals were weak.


H33: Two phones and a Boston onsite with detailed ML follow-ups

Interview dates: February 5 and 12, and March 4, 2020 · Report date: March 11, 2020 · Roles applied for: Quantitative Researcher and QR ML · Location: Boston onsite · Result: Rejected March 9

February 5: First phone

  1. For a Gaussian variable X, find P(X²+2X<3). The candidate’s interval-integration approach was accepted. The report does not give the Gaussian mean or variance.

  2. How can you determine whether a matrix has full rank? What does its determinant mean geometrically? Why is e important?

  3. The candidate chose a CS question over a game-theory question. A hash function returned values from 0 to 3; how could its range grow gradually? The author reported that the interviewer’s proposed answer involved taking modulo 4, and found the wording inconsistent. The original pre-modulo range was not given.

March 4: Boston onsite

  1. After a project presentation with questions from several employees, a probability round asked how many days of equal-probability sampling with replacement are needed to see all five colors. The candidate used geometric waiting times, then was asked to derive the geometric distribution’s expectation.

  2. A paper roll starts with radius R and unwinds at constant linear speed v. Find r(t). The candidate could not derive the model, so the interviewer supplied a differential equation and asked them to solve it. That equation, paper thickness, and core radius were not retained.

  3. In linear regression, what would you do if features are correlated? The candidate suggested PCA to remove correlation, but the interviewer wanted more.

  4. Explain the differences among XGBoost, AdaBoost, and gradient boosting. Then explain how to turn classification into regression. The candidate suggested encoding labels; the interviewer asked for alternatives. The reported direction is classification to regression.

  5. Compare Newton’s method with gradient descent. The candidate mentioned convergence issues for Newton’s method and struggled when asked for further differences.

  6. A sixth coding round had been planned but did not take place.


Keep practicing

In this series

Sources: Anywhere useful interview info can be found.

0 total reactions. Your reactions: none.
0 comments

Comments

0

No comments yet — be the first to share what you know.