In this article, we will tackle three common types of quant interview questions: extracting signal from investment analyst sentiment (Marshall Wace), explaining the role and profits of a market maker (IMC Trading), and solving a geometric expectation problem (WorldQuant). Each section provides detailed solutions, explanations of underlying concepts, and where appropriate, code and mathematical equations.

Quant Interview Questions from Marshall Wace and Other Companies

1. Extracting Signal from Investment Analyst Sentiment (Marshall Wace)

Understanding the Problem

Suppose you are given a dataset of investment analyst sentiment, which might include analyst ratings (Buy, Hold, Sell), target price changes, or written reports. Your task is to extract a predictive "signal" from this data—an actionable piece of information that can anticipate future asset price movements.

Step-by-Step Approach

1. Understand the Dataset
   • What are the fields? (e.g., analyst name, company, date, recommendation, target price, text notes)
   • How frequent are the updates?
   • Are there known biases or patterns (optimism, herding)?
2. Preprocessing and Feature Engineering
   • Standardize Recommendations:
     Map textual recommendations to numeric values. For example:

     Recommendation	Numeric Value
     Strong Buy	+2
     Buy	+1
     Hold	0
     Sell	-1
     Strong Sell	-2

   • Aggregate Sentiment:
     For each stock and date, compute aggregate sentiment, e.g., mean recommendation score, weighted by analyst reputation or track record.
   • Incorporate Target Price Changes:
     Compute the percentage change from previous to current target price for each analyst.
   • Textual Analysis (NLP):
     If you have access to analyst reports, use sentiment analysis (e.g., VADER, TextBlob, or more advanced transformer-based models) to extract sentiment scores from text.
3. Signal Construction
   • Define Response Variable:
     Typically, the response variable is future returns (e.g., 1-day, 1-week, or 1-month forward returns).
   • Regression Analysis:
     Run regressions of future returns on sentiment features to test their predictive power.
     $$ r_{t+1} = \alpha + \beta \cdot \text{Sentiment}_t + \epsilon_{t+1} $$ Where \( r_{t+1} \) is the forward return, and \( \text{Sentiment}_t \) is the aggregated signal.
   • Cross-Validation:
     Use rolling or expanding window backtesting to avoid lookahead bias.
   • Information Coefficient (IC):
     Measure the rank correlation between the signal and future returns.
     $$ IC = \text{corr}(\text{Signal}_t, r_{t+1}) $$
4. Assess Signal Quality
   • Sharpe Ratio: Annualized risk-adjusted return of a strategy based on the signal.
   • Turnover and Capacity: Is the signal stable, or does it require frequent trading?
5. Addressing Biases and Overfitting
   • Analysts may herd or display optimism bias. Techniques such as cross-sectional demeaning or excluding stocks with excessive coverage can help.
   • Regularization (Lasso, Ridge) helps avoid overfitting.

Example Python Code

import pandas as pd
import numpy as np
from sklearn.linear_model import LinearRegression
from scipy.stats import spearmanr

# Suppose df has columns: ['date', 'ticker', 'analyst', 'recommendation', 'future_return']
rec_map = {'Strong Buy': 2, 'Buy': 1, 'Hold': 0, 'Sell': -1, 'Strong Sell': -2}
df['rec_value'] = df['recommendation'].map(rec_map)

# Aggregate sentiment by ticker/date
agg = df.groupby(['date', 'ticker'])['rec_value'].mean().reset_index()
# Merge with forward return
agg = agg.merge(df[['date', 'ticker', 'future_return']].drop_duplicates(), on=['date', 'ticker'])

# Regression
X = agg[['rec_value']].values
y = agg['future_return'].values
reg = LinearRegression().fit(X, y)
print('Beta:', reg.coef_[0])

# Information coefficient
ic, _ = spearmanr(agg['rec_value'], agg['future_return'])
print('IC:', ic)

Key Concepts Explained

• Signal: A quantifiable variable derived from data that helps predict future asset returns.
• Information Coefficient (IC): A measure of predictive power, typically the Spearman or Pearson correlation between a signal and future returns. Higher IC values (closer to 1 or -1) indicate stronger predictive value.
• Backtesting: Simulating how a trading strategy would have performed in the past, using only information available at the time.

2. What Does a Market Maker Do and How Does It Make Money? (IMC Trading)

Definition of a Market Maker

A market maker is a firm or individual that quotes both a buy (bid) and a sell (ask) price for a financial instrument, hoping to make a profit on the bid-ask spread. Market makers provide liquidity to the market, allowing other participants to buy or sell assets quickly.

Role and Functions

• Continuous Quoting: Market makers maintain firm bid and ask quotes for securities throughout the trading day.
• Liquidity Provision: By being ready to buy and sell, they ensure there is always a counterparty for other traders.
• Order Matching: They match buyers and sellers, reducing transaction delays.

How Market Makers Make Money

1. Bid-Ask Spread

   The fundamental way market makers earn profits is through the bid-ask spread. For example, suppose a market maker quotes:

   • Bid: $100.00
   • Ask: $100.05

   If they buy from a seller at $100.00 and sell to a buyer at $100.05, they earn $0.05 per share.

   Action	Price	Profit per Share
   Buy (from seller)	$100.00	-
   Sell (to buyer)	$100.05	$0.05

2. Inventory Management

   Market makers often hold inventory to facilitate trades, but they must manage the risk of price movement. Advanced market makers use quantitative models to hedge inventory risk.

3. Order Flow Information

   By observing order flows, market makers can sometimes infer information about future price moves and adjust quotes accordingly.

4. Rebates and Fees

   Some exchanges offer rebates to market makers for providing liquidity (posting limit orders), which adds to their profits.

Mathematical Example of Market Maker Profit

Suppose a market maker executes 10,000 round-trip trades in a day, each with a $0.02 spread and no inventory risk:

• Total shares traded: 10,000
• Profit per share: $0.02
• Total profit: \( 10,000 \times 0.02 = \$200 \)

Risks Faced by Market Makers

• Adverse Selection: Trading with better-informed counterparties can lead to losses.
• Inventory Risk: Holding inventory exposes the market maker to price changes.
• Technological Risk: Delays in updating quotes or executing trades can lead to losses, especially in fast markets.

Sample Python Code: Simulating Market Making PnL

import numpy as np

# Simulate 10000 trades, each for 1 share, with $0.02 spread
num_trades = 10000
spread = 0.02

# Assume 50% trades executed at bid, 50% at ask
pnl = num_trades * spread
print('Expected PnL:', pnl)

Key Concepts Explained

• Bid-Ask Spread: The difference between the price a market maker is willing to buy (bid) and sell (ask) an asset.
• Liquidity: The ease with which an asset can be bought or sold without affecting its price.
• Inventory Risk: The risk that the value of a market maker’s holdings may move against them before they can offset their position.

3. Geometric Expectation: Expected Sum of Distances in a Triangle (WorldQuant)

Problem Statement

Triangle ABC has sides of length 45, 60, and 75. A point X is placed randomly and uniformly inside the triangle. What is the expected value of the sum of perpendicular distances from point X to this triangle’s three sides?

Restating the Problem Mathematically

Let the triangle have sides \(a=45\), \(b=60\), \(c=75\). For a random point \(X\) inside the triangle, let \(d_a, d_b, d_c\) be the perpendicular distances from \(X\) to sides \(a, b, c\) respectively. Find:

$$ \mathbb{E}[d_a + d_b + d_c] $$

Key Observations and Geometric Concepts

• Area Formula: The sum of distances from any interior point to the sides of a triangle, weighted by the respective side lengths, is constant and equals twice the area of the triangle.
• For uniformly random point inside the triangle, the expected sum of distances to the sides is:
  $$ \mathbb{E}[d_a + d_b + d_c] = \frac{2 \cdot \text{Area}}{s} $$ where \( s \) is the semi-perimeter.

Detailed Solution

Step 1: Calculate the Area of the Triangle

Given sides \(a=45\), \(b=60\), \(c=75\):

• Semi-perimeter \(s\): $$ s = \frac{a + b + c}{2} = \frac{45 + 60 + 75}{2} = \frac{180}{2} = 90 $$
• Area (Heron's formula):
  $$ \text{Area} = \sqrt{s(s-a)(s-b)(s-c)} $$ $$ = \sqrt{90 \times (90-45) \times (90-60) \times (90-75)} $$ $$ = \sqrt{90 \times 45 \times 30 \times 15} $$

Let's calculate step by step:

• \(90 \times 45 = 4050\)
• \(30 \times 15 = 450\)
• \(4050 \times 450 = 1,822,500\)
• \(\sqrt{1,822,500} = 1,350\)

So, Area = 1,350

Step 2: Plug into the Formula

The expected sum of distances is:

$$ \mathbb{E}[d_a + d_b + d_c] = \frac{2 \cdot \text{Area}}{s} $$ $$ = \frac{2 \times 1350}{90} $$ $$ = \frac{2700}{90} $$ $$ = 30 $$

Final Answer

The expected value of the sum of the perpendicular distances from a uniformly random point inside the triangle to the three sides is 30.

Why Does This Formula Work?

• For any point inside the triangle, the sum \( a d_a + b d_b + c d_c = 2 \\(\times \text{Area} \). This is a classic result from geometry known as the "distance sum formula". For a uniformly random point, the expected value of each distance is proportional to the area’s distribution, and when you sum the expected distances, the formula simplifies to \( \frac{2 \cdot \text{Area}}{s} \).
• Intuitively, the result is independent of the specific location of the point, relying instead on the triangle's area and perimeter, reflecting the symmetry of uniform randomness.

Generalization

For **any triangle** with sides \(a, b, c\) and area \(A\), the expected sum of distances from a uniformly random point to all three sides is:

$$ \mathbb{E}[d_a + d_b + d_c] = \frac{2A}{s} $$ where \( s = \frac{a+b+c}{2} \).

This elegant result is useful in both interview settings and geometric probability problems.

Python Code for Verification (Monte Carlo Simulation)

To confirm the analytic result, we can simulate random points inside the triangle and average the sum of their perpendicular distances to the three sides.

import numpy as np

def random_point_in_triangle(A, B, C):
    # Generate random barycentric coordinates
    r1, r2 = np.random.rand(2)
    if r1 + r2 > 1:
        r1, r2 = 1 - r1, 1 - r2
    return A + r1 * (B - A) + r2 * (C - A)

def distance_point_to_line(P, Q, R):
    # Distance from point P to line through Q and R
    return np.abs(np.linalg.det(np.array([Q - R, P - R]))) / np.linalg.norm(Q - R)

# Define triangle vertices (right triangle for simplicity)
A = np.array([0, 0])
B = np.array([75, 0])
C = np.array([0, 60])

# Number of simulations
N = 100000
sum_dist = 0.0

for _ in range(N):
    X = random_point_in_triangle(A, B, C)
    d1 = distance_point_to_line(X, B, C) # to side a (opposite A)
    d2 = distance_point_to_line(X, A, C) # to side b (opposite B)
    d3 = distance_point_to_line(X, A, B) # to side c (opposite C)
    sum_dist += (d1 + d2 + d3)

expected_sum = sum_dist / N
print("Simulated expected sum of distances:", expected_sum)

This code, when run, will yield a value very close to 30, confirming the analytic result.

Key Concepts Explained

• Heron's Formula: Calculates the area of a triangle from its side lengths: $$ \text{Area} = \sqrt{s(s-a)(s-b)(s-c)} $$ where \( s = \frac{a+b+c}{2} \).
• Uniform Distribution in a Triangle: When a point is chosen uniformly at random inside a triangle, every region of equal area is equally likely to be chosen.
• Perpendicular Distance from Point to Side: For a triangle, the perpendicular distance from a point to a side can be calculated using vector cross products or determinants.

Summary Table: Concepts and Formulas Used

Conclusion

The quant interview process at top-tier firms such as Marshall Wace, IMC Trading, and WorldQuant rigorously assesses both analytical depth and mathematical intuition. Extracting signal from sentiment data involves careful preprocessing, thoughtful feature engineering, and robust statistical testing. Understanding the mechanics and risks of market making is essential in today’s electronic markets. Advanced geometric problems, like the expected distance sum in a triangle, test conceptual clarity and problem-solving skills.

Mastery of these concepts not only helps in acing interviews but also lays the foundation for a successful career in quantitative finance. Practicing such questions and deeply understanding the principles behind them is key to building confidence and technical expertise for quant roles.

Related Articles

• Quant Interview Questions from Tower Research Capital
• Susquehanna International Group Quant Interview Questions