I just finished it recently Two Sigma Online Assessment, successfully passed the OA and entered the next round. The Two Sigma Code Test is the most critical first level when applying for jobs at Two Sigma. It is very unique in terms of difficulty and style. Different from traditional OA companies that prefer LeetCode high-frequency questions, Two Sigma's questions focus more on probabilistic thinking, data structure design and engineering implementation capabilities, and the time pressure is relatively high. Below I will share with you the real high-frequency question types I encountered this time, detailed problem-solving ideas and preparation experience.
Two Sigma Code Test overall situation
- Platform: HackerRank (mainstream) / CodeSignal (some positions)
- Question volume: Usually 3 coding questions
- Duration:90~120 minutes
- Difficulty: 1 Easy-Medium + 1 Medium + 1 Medium-Hard
- Features: A high proportion of probability and simulation questions, many boundary conditions, and emphasis on code quality and optimization awareness
The Two Sigma Code Test increasingly examines the thinking ability in actual engineering scenarios. Simply answering questions is no longer enough. You must practice more simulation and probability modeling questions.
Detailed explanation of high-frequency real questions
1. Efficient Regression V5
Question description: Implement linear regression (Ordinary Least Squares) and calculate univariate OLS regression coefficients (betas) between N assets x0~xN and N assets y0~yN. Does not include the intercept term (no intercept). The input is two pandas DataFrame: dfx and dfy, each row corresponds to the return on assets at the same time point.
Problem-solving ideas:
- For each pair (xi, yi) calculate beta = cov(x,y) / var(x) individually
- Pay attention to handling NaN values and data alignment
- Output as beta list
Core code framework(Python):
Import pandas as pd
import numpy as np
def compute_betas(dfx: pd.DataFrame, dfy: pd.DataFrame):
betas = []
for col in dfx.columns:
x = dfx[col].dropna()
y = dfy[col].dropna()
if len(x) == 0:
betas.append(0.0)
continue
beta = np.cov(x, y)[0,1] / np.var(x) if np.var(x) != 0 else 0.0
betas.append(beta)
return betas2.Linear Interpolator
Question description: Given n knot points (x_knots, y_knots), implement a linear interpolation function. For input x_input, the corresponding interpolated value is returned. If x_input is outside the range of knot points, extrapolation is performed.
Problem-solving ideas:
- First sort the knot points by x coordinate
- Use binary search to find the interval where x_input is located
- Compute linear interpolation (or extrapolation)
3. Daily Temperature By Town
Question description: Given multiple days of daily temperature data for P towns and New York City (NYC), answer the following 6 questions:
- Q1: Name of town with the largest temperature change
- Q2: When the temperature in Town2 is between 90 and 100 degrees, the median temperature in NYC
- Q3: Use a simple linear model to predict NYC temperature and calculate the sum of the absolute values of all model coefficients (excluding intercepts)
- Q4/Q5: Find the single or two towns with the most accurate temperature predictions for NYC (based on MSE)
- Q6: Find the five towns with the best joint prediction results
Problem-solving ideas: Use pandas for data grouping, correlation calculation and linear regression fitting.
The OA of quantitative companies is no longer the logic of traditional Internet question answering.
After completing Two Sigma OA, my biggest feeling was not how difficult the questions were, but that the information gap in quantifying the company was really obvious.
Many old interviews on the Internet are no longer of high reference value, especially the Code Test in the past two years. The direction of the inspection has changed very quickly. Many people are still preparing according to the traditional LeetCode approach, but the real exam has become more and more focused on content such as simulation, order book, and implementation stability that are closer to actual trading scenarios.
Later, when I came into contact with Programhelp, I discovered that the high-frequency question types they organized internally overlapped really well with the real exams, including order book, probability simulation, and low-latency implementation. Many of the contents would be covered in advance. Compared with blindly answering questions by yourself, this kind of targeted preparation will be much more efficient.
Also, they will also provide OA practical assistance , simulation training and rhythm guidance, it is quite obvious to help quantitative OA, an exam with particularly strong time pressure. If you are currently preparing for quantitative companies such as Two Sigma, Jane Street, and Citadel, you can check it out.
