Skip to content
Work Practice Products Coding interviews Machine learning coding Quant developer HFT course Blog Method Why me About Enquire Book a call

Practice · Quant prep · G-Research

G-Research interview process

G-Research is a London research firm that hires quants to find patterns in noisy financial data. Its process is mostly technical, with a behavioural round in the middle and a final stage with the people you would work for.

The process, from G-Research’s own pages

  1. Quant research stages. An online application reviewed by Talent Acquisition; a technical interview or remote test; a behavioural interview; further technical interviews; then management interviews with team members, your future manager and functional leadership. Source
  2. Internships. Quant research internships: apply online, an online quantitative aptitude test, technical interviews, and a final interview with a functional lead. The process usually takes one to two weeks. Source
  3. When to apply. The firm says it typically launches its graduate positions in August. Source
  4. The work. Quant researchers look for patterns in noisy, real-world data to predict movements in global markets, with specialisms including high-frequency trading, machine learning, data science and risk modelling. Source

What candidates report

Candidates describe the online test as short multiple-choice questions on mathematics, statistics and logic, and technical interviews on probability, statistics, linear algebra and coding, with more machine learning for machine learning roles.

What G-Research states it pays

G-Research does not state salaries in its adverts; its quantitative research page describes a market-leading package. Comparable pay at firms that do state it is in the Quant Careers Report.

PDF · Edition 2 · £79

The Quant Careers Report

G-Research’s process step by step and the pay it states in its adverts, side by side with 12 other trading firms, plus when to apply and five timed practice papers with worked solutions.

Buy the report — £79 See what is inside

What G-Research is testing

Because the job is finding weak signals in noisy data, the statistics questions probe whether you understand estimation rather than formulas: what changes when you rescale a variable, how biased an estimator is, what a coefficient means on the odds scale. Expect to be asked why, not only what.

The technical rounds are long and deep. An interviewer may stay on one problem for the whole hour, extending it each time you solve a part, so a clean first answer that you can build on matters more than a fast one.

Six practice problems

Written by us to match the types G-Research is publicly reported to ask. None is a copy of any question G-Research has used.

  1. You fit y = a + bx by least squares. Every x is then doubled and the model is refitted. What happens to b and to R²?

    Show the answer

    b halves; R² is unchanged

    The fitted line has to give the same predictions in the new units, so the slope per unit of x halves. R² measures how much of the variance of y the line explains, and rescaling x does not change the fit, so it stays the same. Interviewers ask this to see whether you know which quantities depend on units.

  2. U is uniform on [0, 1]. What is the correlation between U and U²?

    Show the answer

    √15/4 ≈ 0.968

    Cov(U, U²) = E[U³] − E[U]E[U²] = 1/4 − 1/6 = 1/12. Var U = 1/12 and Var U² = 1/5 − 1/9 = 4/45. So the correlation is (1/12)/√(1/12 · 4/45) = √15/4 ≈ 0.968. A clearly non-linear relationship can still have a correlation near 1, which is why a high correlation says little about the shape.

  3. A dense neural network layer maps 784 inputs to 128 outputs, with a bias for each output. How many parameters does it have?

    Show the answer

    100,480

    One weight per input-output pair, 784 × 128 = 100,352, plus 128 biases. Quick parameter counts are a routine sanity check in machine learning interviews, usually followed by a question on memory or overfitting.

  4. You estimate a variance from 5 observations by averaging the squared deviations from the sample mean, dividing by 5. On average, what fraction of the true variance do you get?

    Show the answer

    4/5

    The deviations are measured from the sample mean, which sits closer to the data than the true mean, so they are too small. Their average is (n − 1)/n of the true variance, here 4/5. Dividing by n − 1 instead removes the bias, which is Bessel's correction.

  5. X and Y are independent and uniform on [0, 1]. What is E|X − Y|?

    Show the answer

    1/3

    The density of the gap |X − Y| at d is 2(1 − d) on [0, 1], so its mean is ∫ 2d(1 − d) dd = 1 − 2/3 = 1/3. Equivalently, two random points split [0, 1] into three pieces of equal expected length, and the gap is the middle one.

  6. In a logistic regression, a binary feature has coefficient 0.7. What does that mean for the odds of the outcome?

    Show the answer

    They multiply by about 2.01

    Logistic regression is linear in the log-odds, so switching the feature from 0 to 1 adds 0.7 to the log-odds and multiplies the odds by e^0.7 ≈ 2.01. It does not double the probability: how much the probability moves depends on where it starts.

Practise under time

Written probability test: Fifteen probability and statistics questions, easy to hard, 30 minutes.

How to prepare for G-Research

Revise estimation and regression until you can explain every term, including what happens under rescaling, collinearity and small samples. Practise one long problem per session rather than many short ones, extending it yourself after each step. Have one project ready to discuss in depth, including what did not work and why.

Applying to G-Research? The strategy hour checks your statistics and research story against what its technical rounds look for. What the hour covers, £250.

Other firm guides: Jane Street, Optiver, Citadel, IMC, Five Rings, Two Sigma, D. E. Shaw, DRW, Hudson River Trading, SIG, Flow Traders.