← Morgan Stanley Interview Insights

Morgan Stanley·Data Scientist·Technical Phone Screen·Senior

Senior
Jul 2026

Summary

Market risk and data science screen at Morgan Stanley. Four questions, two of them pretty involved quant problems and two that felt like they were testing whether you actually think or just memorize.

Questions Asked (4)

Q1

Design a stress-testing framework for WTI and Brent crude oil positions, covering marginal modeling, dependence structure, t-copula calibration and simulation, multi-day horizon aggregation, scenario generation (historical, hypothetical, conditional, reverse), and how you'd validate and communicate the framework's limitations.

Technical Trade-offsData ModelingSystem Design
Author's notes

This was the bulk of the interview.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by framing the framework's purpose and scope, then walk through the modeling pipeline from marginal distributions to dependence and simulation, and finish with scenario generation, validation, and communication of limitations. Emphasize practical trade-offs and how you'd ensure robustness and transparency.

Pro tip: Show you understand that no single model captures all risks—demonstrate how you'd combine statistical rigor with expert judgment and stress scenarios to cover tail risks, and always communicate assumptions and limitations clearly to stakeholders.

1. Define Objectives and Scope

Clarify the portfolio, risk measures (e.g., VaR, ES), confidence levels, and horizons. Identify key constraints like data availability and regulatory requirements.

2. Model Marginals and Dependence

Fit marginal distributions to WTI and Brent returns (e.g., using EVT for tails) and calibrate a t-copula to capture tail dependence. Validate the copula fit with goodness-of-fit tests.

3. Simulate and Aggregate

Generate correlated scenarios via the t-copula and inverse marginal CDFs. Aggregate P&L over multi-day horizons, accounting for autocorrelation and overlapping returns.

4. Generate Scenarios

Incorporate historical, hypothetical, conditional, and reverse stress scenarios to complement the statistical model and capture plausible extreme events.

5. Validate and Communicate

Backtest the model, perform sensitivity analysis, and document limitations. Communicate results and caveats clearly to stakeholders, emphasizing model risk.

Key Points to Mention

  • Tail dependence and why t-copula is preferred over Gaussian for crude oil
  • Marginal modeling with extreme value theory (EVT) for fat tails
  • Multi-day horizon aggregation: square-root-of-time vs. overlapping returns
  • Scenario generation: historical, hypothetical, conditional, and reverse stress testing
  • Backtesting (e.g., Kupiec, Christoffersen) and sensitivity analysis
  • Model limitations: parameter uncertainty, regime changes, and communication to non-technical stakeholders

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

Given two normally distributed random variables X and Y with variances and correlation rho, what is Var(X+Y) when they're independent, and when they're dependent? When is X+Y itself normally distributed?

Data Modeling
Author's notes

Felt like a warmup but I still second-guessed myself on the last part.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by stating the general formula for the variance of a sum: Var(X+Y) = Var(X) + Var(Y) + 2Cov(X,Y). Then explain that when X and Y are independent, Cov(X,Y)=0, so the variance simplifies to the sum of variances. When they are dependent, Cov(X,Y) = ρσ_Xσ_Y, so Var(X+Y) = σ_X^2 + σ_Y^2 + 2ρσ_Xσ_Y. Finally, address normality: the sum of two jointly normal variables is always normal, but if they are only marginally normal and not jointly normal, the sum may not be normal.

Pro tip: Emphasize that independence implies zero correlation but zero correlation does not imply independence (except in the joint normal case). Also, mention that in finance, joint normality is often assumed, but it's crucial to verify, as dependence structures can break normality of sums.

1. State the general variance formula

Write Var(X+Y) = Var(X) + Var(Y) + 2Cov(X,Y). This applies regardless of distribution.

2. Apply independence assumption

When X and Y are independent, Cov(X,Y)=0, so Var(X+Y) = Var(X) + Var(Y).

3. Apply dependence with correlation

When dependent, Cov(X,Y) = ρσ_Xσ_Y, so Var(X+Y) = σ_X^2 + σ_Y^2 + 2ρσ_Xσ_Y.

4. Discuss normality of the sum

If X and Y are jointly normally distributed, then any linear combination, including X+Y, is normally distributed. If they are only marginally normal but not jointly normal, X+Y may not be normal.

5. Summarize and connect to context

Relate to data science and finance: understanding dependence and joint normality is key for risk modeling and portfolio variance.

Key Points to Mention

  • General formula: Var(X+Y) = Var(X) + Var(Y) + 2Cov(X,Y)
  • Independence implies Cov(X,Y)=0, so Var(X+Y)=Var(X)+Var(Y)
  • Dependence: Cov(X,Y)=ρσ_Xσ_Y, so Var(X+Y)=σ_X^2+σ_Y^2+2ρσ_Xσ_Y
  • Joint normality ensures X+Y is normal; marginal normality alone does not guarantee it
  • Zero correlation does not imply independence unless jointly normal
  • In finance, joint normality is often assumed but should be validated

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

Classic Monty Hall: three doors, one car, host always reveals a goat after your pick. Should you switch, and what are the probabilities for staying vs switching?

Algorithms & Data Structures
Author's notes

I've seen this a hundred times so I just answered it.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

State clearly that you should switch, then explain the probabilities using a simple enumeration or conditional probability argument. Emphasize that the host's action is not random—it reveals information that changes the probability distribution.

Pro tip: Mention that the host's behavior is crucial: if the host revealed a door at random and it happened to be a goat, the probabilities would be 50-50. This shows you understand the assumptions and can avoid a common pitfall.

1. Restate the problem and assumptions

Confirm the setup: three doors, one car, two goats; you pick one; the host, who knows where the car is, always opens a different door with a goat. This ensures everyone is on the same page.

2. State the answer

Clearly say that you should switch, because switching wins with probability 2/3 while staying wins with probability 1/3.

3. Explain the probabilities

Use a simple argument: initially, the probability your chosen door has the car is 1/3, and the probability the car is behind one of the other two doors is 2/3. When the host reveals a goat, that 2/3 probability collapses onto the remaining unopened door, so switching wins with probability 2/3.

4. Provide an alternative explanation or simulation

If needed, enumerate all possible outcomes (e.g., car behind door 1, 2, or 3) to show that switching wins in 2 out of 3 cases. Alternatively, mention that a Monte Carlo simulation would confirm the 2/3 probability.

5. Address common misconceptions

Clarify that the host's action is not random and depends on knowledge of the car's location. If the host opened a door randomly and it happened to be a goat, the probabilities would be different (1/2 each).

Key Points to Mention

  • Initial probability of picking the car is 1/3.
  • Probability that the car is behind one of the other two doors is 2/3.
  • Host's reveal is not random; it is conditional on knowing where the car is.
  • Switching wins with probability 2/3, staying wins with probability 1/3.
  • The problem is equivalent to being offered the choice between your original door and the other two doors combined.
  • If the host's behavior were random, the probabilities would be 50-50, highlighting the importance of assumptions.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

You have 8 identical-looking balls where exactly one is heavier. Using a balance scale at most twice, how do you always find the heavier ball?

Algorithms & Data Structures
Author's notes

Split into groups of 3, 3, 2.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Divide the 8 balls into three groups: 3, 3, and 2. Weigh the two groups of 3 against each other; if they balance, the heavier ball is in the remaining 2, and one more weighing finds it. If they don't balance, take the heavier group of 3, weigh one against another, and the heavier or the unweighed ball is the answer.

Pro tip: Emphasize that the optimal strategy uses the balance scale's three possible outcomes (left heavy, right heavy, balanced) to maximize information gain, which is why dividing into three groups is key. This demonstrates algorithmic thinking beyond brute force.

1. Understand the problem constraints

Recognize that you have exactly two weighings and need a deterministic method. The balance scale gives three outcomes per weighing, so you can distinguish up to 3^2 = 9 possibilities, which is enough for 8 balls.

2. Divide into three groups

Split the 8 balls into groups of 3, 3, and 2. This leverages the three outcomes of the first weighing to narrow down the heavier ball to a smaller set.

3. First weighing: compare the two groups of 3

Place 3 balls on each side of the balance. If they balance, the heavier ball is in the remaining 2; if not, it's in the heavier group of 3.

4. Second weighing: identify the heavier ball

If the first weighing balanced, weigh the remaining 2 balls against each other to find the heavier one. If it didn't balance, take the heavier group of 3 and weigh one ball against another; if they balance, the unweighed ball is heavier, otherwise the heavier side reveals the ball.

5. Verify and explain

Walk through each possible outcome to show the method always works within two weighings. This demonstrates thoroughness and confidence in the solution.

Key Points to Mention

  • Information theory: each weighing has three outcomes, so two weighings can distinguish up to 9 cases.
  • The strategy of dividing into three groups (3, 3, 2) rather than halves.
  • The decision tree: first weighing branches into three paths, each resolved with one more weighing.
  • The importance of handling the balanced case correctly (remaining 2 balls).
  • Time complexity: O(1) since the number of weighings is constant regardless of input size.
  • Generalization: for n balls with one heavier, the minimum number of weighings is ceil(log_3(n)).

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.