TOC
InternshipCubist：9.5Citadel GQS 1: 9.9CitSec 1: 9.10Citadel Credits: 9.11CitSec 2: 9.19SquarePoint: 9.20Quantbot 10.2BNP 10.3FiveRings 10.4Millennium 10.7-1Millennium 10.7-2Point72 10.11Goldman Sachs 10.11 (ISG)SquarePoint 10.14Verition 10.14Quantbot 10.15-1Quantbot 10.15-2Fiverings 10.15 guangyi yueBnp 10.16-1Bnp 10.16-2Bnp 10.16-3Quantbot 10.18 final roundVerition 10.21ArrowStreet Capital 10.23 first roundPoint72 10.23 second roundIMC 10.25 First roundSquarepoint 10.28 Third roundFiverings 10.29 Second Tech roundIMC 10.31 hr callIMC 10.31 tech round 1IMC 10.31 tech round 2IMC 10.31 round 3Two Sigma 10.7 round 1Two Sigma 11.7 round 2Millennium 11.12 final roundFull TimeBalyasny 9.30 HR callOptiver 9.30 First RoundTwo Sigma 10.1 First RoundVirtu 10.8 First RoundTower — North Moore, Oct. 10, 2025Bam — Gordon Liu, Oct. 13, 2025Point72 — PCAT, Oct. 15, 2025Cubist — Grant Yu, Oct. 15, 2025Optiver — hr call, Oct. 16, 2025JP EMM — intro call, Oct. 21, 2025Optiver r2, Oct. 22, 2025DRW r1, Oct. 23, 2025Tower r2, Oct. 24, 2025Freestone Grove Partners, Oct. 27, 2025Cubist, Oct. 28, 2025Drw r2, Oct. 30, 2025BAM, Oct. 31, 2025Freestone Grove Partners, Nov. 4th, 2025DRW, Nov. 05, 2025Citsec, Nov. 19, 2025Citadel EQR, Nov. 20, 2025Citadel GQS, Nov. 21, 2025SIG, Dec. 1, 2025Headlands, Dec. 2, 2025SIG, Dec. 17, 2025Millennium, Mar. 9, 2026SIG, Mar. 17, 2026R1, QuantR2, QuantR3, TechR4, TradingMillennium, Mar. 18, 2026Point72 IAC, Mar. 19, 2026Cubist, Mar. 19, 2026Dualitas, Mar. 20, 2026Point72, Mar. 23, 2026AQR, Mar. 24, 2026Cubist, Mar. 25, 2026Cubist, Apr. 1, 2026Tower, Apr. 2, 2026AQR, Apr. 7th, 2026Dualitas, Apr. 9th, 2026Point72 IAC, Apr. 10th, 2026Squarepoint, Apr. 13th, 2026Point72 IAC, Apr.15th, 2026Cubist, Apr. 16th, 2026Cubist, Apr. 22nd, 2026Cubist, Apr. 29th, 2026Point72 IAC, May 6th 2026
Internship
Cubist：9.5
Full resume, how to you do single factor backtesting?
Citadel GQS 1: 9.9
过简历，三个项目
how to perform dropout in the machine learning
算法：怎么找一个数列的25th quantile (quick select is the best. Heap is second best)
How to find it online manner? (BST)
CitSec 1: 9.10
resume
Ridge regression: how should we manipulate x1 and x2 in advance
ridge regression: is it always invertible
ridge regression: compare a1, a2, 0, 0.5, 1, where y=a1x1+a2x2+eps, x1=x2(should be 0<a1=a2<0.5)
brain teaser: clock, angle of 12:15
lin reg: corr(X1,Y)>0, corr(X1,X2)>0, corr(X2,Y)=0, y=ax1+eps, y=b1x1+b2x2+eps, compare a and b1 (should be a<b1)
Citadel Credits: 9.11
Resume
What is duration?
What is decorator in python?
Lin reg: y=ax+eps, x=by+eps, relation between a and b
CitSec 2: 9.19
Resume
Two portfolio, mu1,sigma1, mu2, sigma2, want to maximize sharpe, how to allocate
Create a dataframe
SquarePoint: 9.20
Resume, vwap and NLP
Coding: queue at ATM (should find number it occurs first, then sort and give results)
Coding: number of subarrays smaller or equal then k (should use double pointers)
Brain teaser: why HHHH is harder to get then HTHT. (because of overlapping.)
Calculate Expected number to get HHHH
Stat: fair coin, throw 100 times, 55 head, possibility. (z score)
Lin reg: double the data, what will happen to r2, t, beta. (t will devided by sqrt(2))
Quantbot 10.2
Pure Behavior Question:
How many experience, why do you want to do quant, what part of quant is it that you like the best, why do you want to work here, what do you wanna get here.
BNP 10.3
Resume related:
Future factor digging and stock factor digging difference? TA-Lib how to construct factor?
Introduce time-variation of Transformer
C++ & Python difference (compiled & explanatory; running speed; memory control)
What is static method? (belong to class rather than instance, can’t modify non-static value)
Regression: var(x)=10, var(y)=20, y=ax+eps, a=1, x=by+eps, what is b? (1/2)
Stat: keep rolling a fair dice until 6 appears, what is expected number of 5 occurred? (1)
FiveRings 10.4
Just explaining the process.
Millennium 10.7-1
General Resume Question
How to elaborate a technical term to someone who is not in this field? (Visual, Analogy, …)
Your proudest achievement?
Millennium 10.7-2
Resume Question（NLP project，Commodity Futures）
A coin, 10000 toss, 5200 tail, test if it’s fair ((5000-5200)/sqrt(10000*0.5*0.5)=4>1.96, unfair)
If we have a linear regression with beta0, what is sum of residuals? (0, because can move beta0)
X,Y normal dist. With correlation coef rho, E(max(X,Y))?
Point72 10.11
Resume Question:
Explain more on the Markov chain model, what are the states?(mid-price movement, percentage, past 10 trading time movement to forecast next price movement)
Transformer model: why use transformer model, compare with LSTM (self-attention, have a better long-short portfolio return than LSTM, every two weeks to rebalance the portfolio)
Factors, what are they, are they technical factor(Yes, price, volume, order book data)
Goldman Sachs 10.11 (ISG)
Introductory Call. Explaining the process to me.
SquarePoint 10.14
Suppose I have standardized x and y, what’s the relation between coef. Of X~y and Y~x(same)
Derive the function of beta hat in matrix form (XTX)-1(XT)Y
Coin 100 times, 56 head, if fair (1.2, compare with 1.96)
MLE for exponential distributions(1/mean), why choose MLE (asymptotic efficiency)
2% return in 2year, 3%ret in 3 year, 4%ret in 4 year, which is preferable (2% one, need to find (1+0.01x)^(1/x) take derivative is monotonic decreasing)
Sharpe ratio of 1, prob. of loss money after 4 years. (z=4sigma/2sigma=2, (1-95.5%)/2=2.25%)
Verition 10.14
Basically full resume question
How to calculate IR from IC? ( IR = IC * sqrt(N))
Suppose I have a in-sample factor, I tested 10^5 different feature and the highest one has a t-value of 4, and on this one the out-of-sample t-value is also 4, what can you say about implement this in the real market? (The expected t-value in production should be smaller than 4, )
16factor, suppose each of them is sharpe 1, what’s the 16 factor combined sharpe? (if no correlation, 4, if with 10% correlation, around 3.5.) I achieved a 2.8 Sharpe, means my correlation is around 15%.
Why in Transformer usually use layer_norm, not batch_norm? ()
How to construct an order book (what kind of api: order issue and trade data to update my orderbook)
What’s the data structure for an order book if implement by yourself? ()
Quantbot 10.15-1
Full of tech questions.
Formula of beta hat in matrix form, Geometric meaning of beta hat (orthogonal projection of Y on columns of X), geometric meaning of R2 (cos2, angle of Y and Y hat), geometric meaning of t value (t=cot(theta)*sqrt(n-2))
Duplicate data, change in beta, r square, t-stat
Y=aX, X=bY, is a and b same? (No, one is for residual of y, one is for residual of x)
X and Y iid. N(0,1), P(X>3Y) (1/2, just draw a line)
Suppose n rv, each have rho, what’s the range of rho (-1/(n-1) to 1, by calculating the det of matrix)
Give me some factors (reversal, difference between exchanges)
Tell me a recent book/ news you read that has an impact on quant finance (Policy change and Federal rate cut on Chinese A-share stock market, and how to do risk management.)
Quantbot 10.15-2
Resume questions.
Order book description
Transformer structure
Suppose I have 5 years of data, and 100 features, create a strategy (remember to cut the time into training, validation and test).
Fiverings 10.15 guangyi yue
General Setup: a frog jump, n stairs
Q1: if not restrictions, total possible ways of frog going up?
Q2: if can only have odd number of jumps, total possible ways of frog going up?
Q3: if every jump must be odd, total possible ways?
Q4: if number of even jumps is even, total possible ways?
Q5: relationship between Q2 and Q4?
Ans (1. , 2. , 3. Fib(n), 4. , 5. When N is odd, complement, when N is even, same)
Idea of solving this:
Q2, thinking about only the N-1th slot, determined by prior ones
Q3: recursion
Q4: mapping, for every good one, change the first stair from x to 1-x, is the disqualified one
Q5: just counting the number of 1s and 0s in the sequence.
Bnp 10.16-1
Market news. What are you tracking (semiconductors, ai)
Tell me more on Chinese A share stock market.
A simple question on BS model (a stock with no volatility, r=5%, S=100, T=1, c=?)
If I ask you to construct something, what is your approach? (You need to answer that you will ask around first to see if this feature has been done by other people first.)
Bnp 10.16-2
Behaviour.
What do you think about the regulations on the financial market, are they necessary.
The one person you admire the most.
Do you think AI will substitute human in finance industry
One time you have a disagreement with people in your group.
Proudest achievement.
What can you contribute to the team.
Do you think today’s fed rate is higher than 1980s? (No, peak is 20%, to control inflation)
Bnp 10.16-3
An insurance kind of question. Define a default rate, I pay quarterly coupon C, receive K if something happens. Relationship between C, K, and default rate lambda.
Quantbot 10.18 final round
Mainly just casual discussion.
If the close price of today is just above a round integer, what is most probably gonna happen tmr? (look at it as a barrier or resistance, more likely to continue going up.)
C++ related:
struct S{
double d;
bool b;
int I;
char c;
}
How much memory does it take? (4*8 or 3*8 depend on complier)
If realign it, how much memory? (16)
What if I add a virtual function in it? (virtual void func(), need a virtual pointer, which is 8 bytes)
What if I add an actual function in it? (Not change, Adding a non-virtual member function does not affect the memory footprint of each struct instance. The function code exists separately in the program's code segment.)
Verition 10.21
Regarding the data project, discuss for 1 hour and then perform for 3 hours.
ArrowStreet Capital 10.23 first round
Resume question
The llm model, why don’t just use the binary label of whether it is halted or not?
Factor backtesting, how did you do that? (cumulative return, and also average return, see stratification)
60 students, into 3 different groups, 20 each, Tom and Jerry prob. of getting into same class? (19/59)
Lin reg: univariate regression, beta hat, with or without interception.
Attenuation bias
What is the equation of standard error?
If with correlation in eps, what direction will it be? （depend on the autocorrelation actually）
What is beta in finance?
What is market beta?
How to derive the beta? (I answered just draw a tangent line)
Point72 10.23 second round
Resume related question on every project.
Suppose Given a data from Yahoo Finance, want to predict future 2 week return.
How to do EDA?
How to deal with missing value?
How to deal with outliers?
How to construct features? ()
Why categorize into different groups （prevent high correlation between factors）
I have x,y,z, corr(x,y)=corr(x,z)=0.9, range of corr(y,z)
IMC 10.25 First round
Question 1
How many path from NY to SF (10C5)
How many path from NY to SF given need to pass through NS (5C2*5C2)
How many path from NY to SF given cannot pass CH (10C5 – 4C1*6C2)
How many path from NY to SF
Consider the two yellow dots: 5C2*5C2 + 5C2*5C2
Question 2.1
New York, San Francisco, Nashville
50% population from New York will leave the city
50% population from San Francisco will leave the city
100% population from Nashville will leave the city
The probability to travel to other cities is the same
What is the long term distribution of population?
Markov Chain, New York: 40%, San Francisco 40%, Nashville 20%
Question 2.2
Same conditions, in addition
People who travel to New York will remain in New York
People who starts from New York need to travel to another city then come back to New York
What is the expected # of days for everyone to end up in New York
Question 2.3
Suppose 2 people, 1 with exp. Lambda1, 1 with exp. Lambda2, each 50% probability, what is the mean and variance of the expected time of arrival?
Coding Question
Given grid matrix 9*9, contains scenic value for each grid
You can only move right and down,
Find the maximum scenic value
- What if have to pass (5,4), and cannot pass (1,4)?
Squarepoint 10.28 Third round
Linear Regression: explain each assumption. What is BLUE?
Explain lasso and ridge. Why is lasso dragging all x down in the same level? Use geometric way to explain it.
Explain lasso and ridge in terms of bias and variance tradeoff.
Choose x randomly in [1, m] with put back n times, the expected number of getting selected. (just like cross validation.)
I have a sequence, Sharpe is S, suppose half of the days have 0 return, delete them, what’s the new Sharpe? Relationship?
Suppose I have a song list, but this list only gives 2 options, I can either do next or I can do shuffle (the song list sequence is fixed, when I shuffle just make me to a random song). Assume start from number 1, want to go to T, what should be my strategy (a threshold strategy, suppose N song, threshold is D, solve for the optimization problem.)
Fiverings 10.29 Second Tech round
Suppose I have a fair coin, I throw it n times. A consecutive heads or tails is a block. Eg, HHTTH have 3 blocks.
- I want to find the expected number of blocks. ((n+1)/2)
- Expected length of the average of blocks. (2^n-1/2^(n-1))
IMC 10.31 hr call
Why IMC
Explain the work I did at uChicago
One time deal with setbacks
One time have a disagreement with someone on the team
IMC 10.31 tech round 1
Three people, decide whether to go to a certain restaurant. They have the same going probability
F(n) = 1-(n/3), for n=1,2,3; and F(n)=0 if n=0
Where n is yesterday the number of people go there.
- Given day 0 n=2, what is the expected number of day to arrive to 0
- Given n=1, and F(0)=1or2, each with 50% probability. Long term expected number of people coming to the restaurant every day.
Suppose in total 2 clients, at every day 50% only one shows up, 50% 2 of them show up.
- Probability that exactly 4 time people cumulative show up?
- Probability that exactly 400 time people cumulatively show up?
IMC 10.31 tech round 2
Linear programming.
First question is a Lagrange multiplier with only one constraint, just use this.
Second question is max x+0.8y+2z with 4 different linear constraint, and the way should be it’s most probably on a vertice.
IMC 10.31 round 3
- Secretary problem.
- Given temperature and length of queue at a playground, they appear to have a lagged relationship, how to build a model to predict the length.
Two Sigma 10.7 round 1
Coding:
We have a collection of temperature and humidity readings from multiple cities. We want to combine these records together by joining each temperature reading to the most recent humidity reading in the same city, as of the time of the temperature reading. Both data series are represented as (time, city, reading) triples and are known to be sorted by time.
The fastest way should be O(1) time.
我的方法是每次二分查找。但由于本身是有序的，可以直接按顺序查找。
Two Sigma 11.7 round 2
Data analysis:
Predict New York housing price.
Millennium 11.12 final round
- T test on sharpe ratio.
Look on chatgpt to see how to derive this.
- Call option, K=0, S is standard normal, what is the call’s price. (integral)
- I have a pandas data frame, with 1 million rows and 3 columns, y, x1, x2. Fit a linear model of y using x1, and get a r square close to 0. Run regression y using x2, also the r square close to 0. Now I run a regression of y over x1 and x2, get a r square close to 1. Is this possible? (It is, use geometric meaning.) Construct three sequences. (y is just a random sequence. X1 is y+noise, where noise is random sequence with large scale, and x2 is y-noise).
- Behavior: Why qr. If have an offer from google, will you go there.
Full Time
Balyasny 9.30 HR call
Why I want to come to US. Experience in the summer. Top 3 most valuable point I would like when I choose company (Exposure, Collaboration, Freedom of operations).
Location. Offer from summer (I say I declined).
Optiver 9.30 First Round
对于card draw，不同的market情况下你的投注概率分布是什么？concave 还是convex
Two Sigma 10.1 First Round
如何选择买一家餐馆？任选数据。
Virtu 10.8 First Round
Resume，critical feedback from last internship，asset class.
Brain teaser: clock的追及问题
Tower — North Moore, Oct. 10, 2025
Resume:
MLP
explain what you did for the first project (de-event volatility), two different ways. Whole process: dataset —> how to de-event —> how to use this for further project.
Skewness of the event vol impact: given 30-delta, 70delta, before and after event, what’s skewness change/ measure of this.
How to avoid overfit in my process: Prior (intuition), rolling -train, only select hyperparameters once.
Dacheng Xiu project, introduce.
Idea and intuition behind how to improve the execution implementation shortfall.
How to weight between liquidity(first level) and fill rate.
Bam — Gordon Liu, Oct. 13, 2025
Lin reg: x, y iid, y ~ x beta? now add moving average, , what’s the autocorr of x? what’s comparing with original beta? What’s new SE(beta)? what’s new t stats?
100人来租房，每个人有endowment，我依次报价，只要报价低于第i个人的endowment就卖给他，然后结束，否则进行到下一个人，继续报价。请问怎么报价能实现利益最大化？每个人的价格分布都是Unif[2000, 5000]
Point72 — PCAT, Oct. 15, 2025
Background info. Intro of last two internships.
recent market news
what’s something I need to improve on
what’s something my peer does better than me
Cubist — Grant Yu, Oct. 15, 2025
Fully resume. mlp project talked for 70min.
Optiver — hr call, Oct. 16, 2025
Why interested in trading industry.
Why want to go optiver
Most challenging project
If want to redo anything in internship project, what would you do
JP EMM — intro call, Oct. 21, 2025
Market making, research oriented, no manual risk, run maintainance.
Research on how to set the spread, etc.
1 microsecond
Optiver r2, Oct. 22, 2025
DRW r1, Oct. 23, 2025
Andy Chen, global macro vol desk
Resume related, on vol and my project.
Linear regression: derive OLS beta. If 150 features and 2000 observations how to deal with this high correlation (I said drop features, ridge, PCA).
Explain PCA. How to select number of features kept in PCA? I said look at the plot of total variance explained to number of features, use elbow point. Why PCA might perform bad in regression?
What should we test on the residual from a regression? I said E(u)=0, and should follow a normal distribution, otherwise the t-stat is not reliable anymore.
If a model performed unrealistically well, what might be the reason? I said if time series then lookahead bias, other possible explanations is overfitting, in which case I should use adjusted R2 then R2.
Tower r2, Oct. 24, 2025
Trent
On volatility surface related questions.
How did I calculate the event volatility? 2 different ways. Explain them. How to compare which model is better, without using the whole pipeline? (I said the model should be smooth and have a temporal stability, he said compare with the actual realized vol).
Then, suppose I have a model to predict the realized vol, and it’s pretty accurate, and I know the rvol two month from now is higher than implied vol, how should I trade this? I said long a straddle, he said need a better answer, I then said swap, he asked how to create this swap. The answer is short a short-term straddle and long a long-term straddle.
Then, how should I hedge the risk? Vega control, delta hedge. Vega control means the weight I short and long the straddles should be decided by vega.
Freestone Grove Partners, Oct. 27, 2025
Natalia
Introduce the team: fundamental discretionary trading, data scientist work as quantimental.
Introduce about past experience, why come to this program.
What’s something like and don’t like in past experience.
Work me through the whole process of doing research, did you gather the data by yourself or provided.
Cubist, Oct. 28, 2025
Jeremy Woo
resume.
What is robustness of ridge regression? Suppose you have many X which are highly correlated, what would the sum of new betas compare with only one x, compare with OLS?
do you need to normalize the input prior to input to ridge?
Drw r2, Oct. 30, 2025
Resume. Why do you say economic intuition is important? Where did you show this?
2iid RV, positive numbers, z = x/y, E(x/y)>1 or ≥1? I said Jensen Inequality, strict greater.
Linear regression: y～x. now (y+u)~x, beta change? CI change? now y ~ (x+u), beta change? CI change?
Heteroskedasticity: how to deal with that? I said HAC, WLS. What is WLS? explain. Is WLS Blue? I said yes, if you know the distribution of . What if you don’t know distribution? Don’t know
On a scatterplot, OLS line, and first PCA line, is that the same?
2 time series, x and y, suppose OLS r2 big, what to do next? Don’t know
Suppose monte carlo to gen a rare event, how to get a good estimation? Not sure….
What’s the difference between AR1 and Random walk? What’s variance of RW? I said t*sigma2, blow to inf.
Suppose you need to build a big project, how to approach this? In C++, what steps should you do? I said parent class. Public api.
BAM, Oct. 31, 2025
Manu.
How did you weight different single name after you have the signal for every single name in your universe every day? （sqrt of market cap)
Coding problem: Permutation(Leetcode)
Coding problem: Optimal Liquidity problem:
Ans:
# b*P_i+1 -(a+2b)*P_i + b*P_i-1 = 0 import math def get_path(a,b,P0,PT,T): A = b B = -a - 2*b C = b if b == 0: # avoid division by 0 path = [P0] + [0] * (T - 1) + [PT] return path # need to solve for root of: Ar^2 + Br + C = 0 Delta = B*B - 4 * A * C if Delta < 0: cos_t = (a + 2*b) / (2*b) # r1 * r2 = A/C = 1 rho = 1 # solve for theta theta = math.acos(cos_t) # P0 = C1 # PT = C1 * cos(T*theta) + C2 * sin(T*theta) C1 = P0 C2 = (PT / (rho ** T) - C1 * math.cos(T*theta)) / \ math.sin(T * theta) path = [(C1 * math.cos(i * theta) + \ C2 * math.sin(i * theta)) for i in range(T+1)] else: sqrt_D = math.sqrt(Delta) r1 = (-B + sqrt_D) / (2 * A) r2 = (-B - sqrt_D) / (2 * A) # now we have r1, r2, need to solve boundary condition: if r1 == r2: # P_t = (C1+C2*t)*r1**t # P0 = C1 # PT = (C1 + C2*T) * r**T C1 = P0 C2 = (PT/(r1**T) - C1) / T path = [(C1+ C2*i)*(r1**i) for i in range(T+1)] else: # normal # P_t = C1*r1^t + C2*r2^t # P0 = C1 + C2 # PT = C1*r1^T + C2*r2^T C1 = (PT - P0*(r2**T))/(r1**T - r2**T) C2 = P0 - C1 path = [(C1*(r1**i) + C2*(r2**i)) for i in range(T+1)] return path if __name__ == "__main__": a = -1 b = 2 P0 = 10 PT = 0 T = 20 path = get_path(a, b, P0, PT, T) print(path)
Freestone Grove Partners, Nov. 4th, 2025
Introduction to the firm and the role mainly.
DRW, Nov. 05, 2025
S~Normal(100, 20), K=120, 估算call price
Citsec, Nov. 19, 2025
1 brick, 3 part, E[longest part]
怎么在linear regression里面做feature selection？前项选择/后项选择。
Citadel EQR, Nov. 20, 2025
Y~N(X, 1), X~N(0,1), what’s Y? (Y=X+eps —> Y~N(0,2))
Y~N(0, X^2), X~N(0,1), what’s Y? (I just said it’s a heavy tail dist.)
regression:
Y~X, now X_new = Sum_{past 5 days of X}, what will change?
Y~X, now Y_new = Sum_{past 5 days of Y}, what will change?
Citadel GQS, Nov. 21, 2025
Easy way is to just use a dict to save prices, and then search prefix.
Can also use a Trie.
- Stock buy and sell, 1 and 2 times
- Newton’s method for solving the sqrt of x
- monte carlo for pi
- random pick 3 numbers without replacement from 1 to 20, prob. that one is average of others
- If X, Y independent, is min(X,Y) and max(X,Y) dependent?
- Explain the difference between LASSO and ridge
SIG, Dec. 1, 2025
What’s your hobby? (Poker)
3 iid. r.v. Uniform (0,3), probability that median between [1,2]
5 dice with 1-9. Now 2,2,4,6,7. Winning condition: aaabb/ a(a+1)(a+2)bb. Like majhong. Each time reroll one. what’s the best strategy? (always reroll 4, E=9/2=4.5)
Follow up: 24446 and 22246 which has a smaller expected winning time. (we both reroll 6. first can change to 24445, then reroll 2 can have 3 possible way: 34445,44455,44456; whereas second one can be 12224 in symmetry, but can only reroll 4 to be 11222,12223, can’t be 01222. Therefore, 24446 has a shorter expected time).
Headlands, Dec. 2, 2025
lin reg assumptions
Feature selection methods （lasso, ridge)
y~x1, r2=0.2, y~x2, r2=0.2, y~x1+x2, r2? (0.2 - 1)
y~x1+x2, r2 v.s. y_hat1=x1+eps, eps_hat=x2+u, yhat=y_hat1 + eps_hat, which one r2 big?
SIG, Dec. 17, 2025
y=beta*x + eps, x input, y output. x between -1 and 1. You can select 24 input, and observe 24 output, to calculate beta. What’s the best way to select? (min var(beta_hat), should choose 24 value on either -1 or 1). Why is this better than just choose uniformly on -1 1? I said signal-noise ratio.
A, B, C selection problem.
streaming data find median, input only between 0 and 10 integers. Algo should be using a counter rather than two heaps.
Millennium, Mar. 9, 2026
Talk about experience at MLP
question: some about a summation over 0 to x, simple
If I have a regression, using 5 features, daily freq. to predict the return of the next 3 month, what might be the issue? (High autocorrelation, very little independent observation)
SIG, Mar. 17, 2026
R1, Quant
derive put call parity
A card game, have 10 card, from A to 10, face down on the table. I have initial value of 1, each time I call a number, then flip 1 card sequentially. If the number I call equal to the number of the card, I multiply the number, otherwise stay unchanged. What’s the best strategy, and what’s the EV of the game?
R2, Quant
talk about experience at Millennium.
suppose I have a product, value to me is x ~ uniform [0, 1m], I know this value. To you, it’s worth y =1.5x. But you don’t know the value at the beginning. One time auction, if you bid the value greater than x, I sell this to you, and reveal the value x. What’s the bid you should call? (Should be 0, winner’s curse)
X ~ N(0, sigma^2), I observe 1 value k=2, what’s point estimation of sigma? (2) What’s the 90% confidence interval of sigma? (highly skewed)
R3, Tech
element wise addition, v.s. numpy addition
a groupby & transform simple question
for this question, I used a interval to save the satisfied time of each symbol.
self._interval = defaultdict(int)
on every trade, if for this symbol I have past # of threshold trades within the lookback period, I set the self._interval[symbol] to be the first time of the last threshold trade + lookback.
In query, I only need to loop over the interval and add them together.
R4, Trading
Play a poker game, in between.
The key is that if using kelly criteria, what’s the initial endowment you have. This is set by you.
Millennium, Mar. 18, 2026
A lot of poker probability, like pair of cards is a pair, is 4 of a kind, etc
calculate 2^20 by mind.
German tank problem.
Suppose in the past 10 years the prob. of location X raining is 30%. I went there for a 8 day trip, and in the first 7 days it doesn’t rain, what’s the prob. that it rains on the last day? (I said might be seasonal, highly unlikely to rainly)
A game theory problem, with very bad description.
Point72 IAC, Mar. 19, 2026
Xuan Yang
primarily trade index - single name dispersion.
different between carry pnl and vega pnl. Why don’t just use IV-RV as target.
What’s typical market behavior for Implied Earning Vol (selling earning vol make positive pnl).
How did you calculate turnover? Is this correct? (just holding the 3m atm straddle every day require turnover, so not the same as equity)
What’s the difference between trading option and stock?
suppose have a call with S=K=100, c=4, r=0. What’s the price for a put with S=100, K=99? (put call parity and delta ~ -0.5)
Expected time to get a HH for fair coin.
The game is, I win if get HH, you win if get HT, is this fair?
Simple linear regression, y~x and x~y, is this overlapping? (beta not the same)
Simple linear regression, double the data, what happen to line and the p-value.
Cubist, Mar. 19, 2026
Harshaan Dargan
Affect of interest rate increase to stock price (from business perspective, and investor perspective)
If Japan increase it’s interest rate, what’s gonna happen to spot exchange rate to US.
Why is gold price increase in past few months.
A few lin. reg. questions: assumptions, double the data, how to deal with multi-collinearity
Suppose given a data of daily temperature across different zip code in U.S., how to use this to create features, and what target you can use this data on.
Matrix size x,y, fill each cell using numbers 1 to N. Can be repeated, sum of each row and column should be divisible by N. How many ways of fillling ()
break a stick into 2 pieces, What’s E(shorter/longer)? ()
Dualitas, Mar. 20, 2026
What’s transpose PCA, is it better at explaining the features.
Vol surface: why there exists a vol surface skew? What’s the distribution of log return in reality.
If you want to create a vol surface, should you use all the strikes
Point72, Mar. 23, 2026
Yilun
On my resume with very detailed question.
How did I create the vol surface. Why use different K.
In the backtesting, did I change the dollar volume roughly the same every day (change of position)
Lasso, is it a quadratic? (No, has abs value). How to change it to a quadratic solver with constraint? (change abs(beta) with u, st. u≥beta, u≥-beta, u≥0). Prove it.
AQR, Mar. 24, 2026
James
general question on resume and interest
how to do textual data sentiment analysis
how to use intraday data to get daily feature
what’s the difference between research on macro and equity
Cubist, Mar. 25, 2026
Emma
Resume question on using announcement data to predict high risk stocks, what type of data used, what kind of model, how to deal with false positive.
Pandas question:
1 data frame, index date, column 100 stock, value is daily return, want to get another matrix, which is distance of last sign change, eg. Days since return flipped sign (+ to - or - to +)
import pandas as pd import numpy as np df = pd.DataFrame( columns=[f"stock_{i}" for i in range(100)], index = pd.date_range(start="2020-01-01", end="2020-12-31", freq="D"), data=np.random.normal(loc=0, scale=1, size=(366, 100)) ) signs = np.sign(df) changed = (signs!= signs.shift(1)) & signs.notna() & signs.shift(1).notna() cnt = pd.DataFrame( columns=df.columns, index=df.index, data=np.arange(366)[:, None] * np.ones((1, 100)).astype(int) ) last_one = cnt.where(changed).ffill().fillna(0).astype(int) new_df = cnt - last_one + 1
How to generate systematic trading signal of oil prices (futures), think of what type of sources you will look at, the logic behind it, the features you want to build off it, and how to find the data.
Cubist, Apr. 1, 2026
Sebastien Penet
Mostly resume questions. What type of pnl, regarding features, does vrp have seasonality, what other features. Ridge regression.
Greeks: what is delta, what is vega, what is vega decay.
What is ATM option and ATM forward.
how to do ridgecv.
Tower, Apr. 2, 2026
resume related. ML
AQR, Apr. 7th, 2026
Fred Liu
Ridge: 当penalty特别大的时候，后果是什么？（想要的答案是会把所有beta都drag到接近0的地方，relative beta变成接近1，本来有信号的feature变成和noise贡献一样的权重了）
Dualitas, Apr. 9th, 2026
Michael Liu
Everything on my resume.
Linear regression assumptions, R2, correlation easy question, merge in pandas, loc and iloc in pandas.
Point72 IAC, Apr. 10th, 2026
Xiaowei Tan
Regarding Experience at MLP
what is no arbitrage condition of the vol surface?(No calendar spread, no butterfly spread, etc)
Suppose 2 call options, same K, T1 < T2, if c(K, T1) > c(K, T2), construct arbitrage.
Squarepoint, Apr. 13th, 2026
Matt Smith
Coding: matrix, longest increasing path. (dfs + memo)
Prob.: throw 2 coins, each time gain 1 dollar, if in the whole history observe both HH and TT then clear all gains. (eg suppose sequence is TH, TT, HT, TT, HH, then the total endowment is 1,2,3,4,0). What is best strategy and what is expected value.
Stats: Suppose I have a series, with half of them = 0, the other half with (mu., sigma), what’s gonna happen to sharpe if I delete all 0s from the series. (total variance formula)
Point72 IAC, Apr.15th, 2026
superday
Basically resume related question.
Some tech:
how to load in large data (have I tried polars before)
random choose 3 points on a circle, prob. that the triangle covers the center
Suppose I have a dict like {’NYC’: 8e6, ‘LDN’: 10e6, ‘HK’: 7e6}, I want to create a class with a function sample, that returns the city name by it’s probability.
There’s a warning in python if you give a default value to the initialization function of a class, what is that?
Cubist, Apr. 16th, 2026
Flavien
Resume question, suppose you have the vol surface dataset, give me some features that you think might / might not work on predicting underlying stock returns.
suppose 300 stocks, daily return N(0,1), what’s the expected number of series that in 5 years Sharpe > 2?
Cubist, Apr. 22nd, 2026
Gregoire Lachaise
Resume question
coin, 60 head, 40 tail, p-value
Cubist, Apr. 29th, 2026
Luv Iyer
recent news
difference between us fed and Europe central bank.
influence of fed rate cut to stock, bond, oil, fx
Point72 IAC, May 6th 2026
Taylar Ma