blog-cover-image

Graviton Quant Analyst Interview: Key Questions and Answers

Adnan

Aug 15, 2026

Landing a quantitative analyst (Quant) role at Graviton Research Capital is highly competitive, with interviews designed to rigorously test your understanding of mathematical finance, statistics, and linear algebra. In this article, we solve and explain key questions you might encounter for a Graviton Quant Analyst position, focusing on covariance matrix ranks, Markov chains, and correlation effects. Each section provides not just the answer, but also a deep dive into the underlying concepts, mathematical reasoning, and their practical interpretation.

Graviton Quant Analyst Questions – Solved & Explained

1. You have returns of 100 stocks, but they are all generated by exactly 5 independent factors with no idiosyncratic component. What is the maximum rank of the covariance matrix?

Understanding the Question

Suppose you have a dataset of returns for 100 different stocks. Each stock's return is driven solely by 5 independent risk factors—there are no idiosyncratic (stock-specific) risk components. You are asked: What is the maximum rank of the covariance matrix of these 100 returns?

Key Concepts Involved

Covariance Matrix: A square matrix where each element \( \Sigma_{ij} \) represents the covariance between the returns of stock \( i \) and stock \( j \).
Rank of a Matrix: The maximum number of linearly independent rows or columns. For a covariance matrix, the rank indicates the number of independent sources of risk that drive the returns.
Factor Model: When returns are generated by a set of factors, the covariance of returns is determined by the covariance of those factors.
No Idiosyncratic Risk: Means that there are no unique risk components associated with any stock; all risk comes from the factors.

Mathematical Explanation

Let's formalize the returns of the 100 stocks. Let \( r \) be a \( 100 \times 1 \) vector of returns, and \( f \) be a \( 5 \times 1 \) vector representing the 5 independent factors. The return for each stock can be written as:

\( r = B f \)

Where \( B \) is a \( 100 \times 5 \) matrix of factor loadings (how sensitive each stock is to each factor). Since there is no idiosyncratic component, this model is exact.

The covariance matrix of returns, \( \Sigma_r \), is:

\( \Sigma_r = \mathbb{E}[ (r - \mathbb{E}[r]) (r - \mathbb{E}[r])^T ] = B \Sigma_f B^T \)

Where \( \Sigma_f \) is the \( 5 \times 5 \) covariance matrix of the factors.

What is the Rank of \( \Sigma_r \)?

\( B \) is \( 100 \times 5 \), \( \Sigma_f \) is \( 5 \times 5 \), so \( \Sigma_r \) is \( 100 \times 100 \).
The rank of \( \Sigma_r \) cannot exceed the rank of \( B \) or \( \Sigma_f \).
Since the factors are independent, \( \Sigma_f \) is full rank (5).
If \( B \) has full column rank (5), then \( B \Sigma_f B^T \) has rank at most \( 5 \).

Therefore, the maximum rank of the covariance matrix of the stock returns is 5.

Intuitive Explanation

All 100 stock returns are just linear combinations of 5 independent factors. This means that, in the 100-dimensional space of returns, they all lie within a 5-dimensional subspace. There are only 5 sources of randomness or risk, so all covariances can be explained by these 5 factors. The other 95 dimensions are just linear combinations and do not add new independent directions.

Equation Summary

\[ \text{Let } r = B f, \quad \text{with } f \in \mathbb{R}^5, B \in \mathbb{R}^{100 \times 5} \] \[ \Sigma_r = \mathbb{E}[r r^T] = B \Sigma_f B^T \] \[ \text{rank}(\Sigma_r) \leq \min\left( \text{rank}(B), \text{rank}(\Sigma_f) \right) \] \[ \text{rank}(\Sigma_r) = 5 \] \] \text{(if both \(B\) and \(\Sigma_f\) have full rank)} \]

Practical Implication

If you try to invert the covariance matrix for portfolio optimization (e.g. in Markowitz's mean-variance optimization), you will find it is singular (not invertible), since it is not full rank. This is a common issue in factor models without idiosyncratic risk; regularization or adding small noise ("shrinkage") is often required.

2. A Markov transition matrix has eigenvalue 1. Why?

Understanding the Question

A Markov transition matrix describes the probabilities of transitioning from one state to another in a Markov process. The question asks: why does this matrix always have eigenvalue 1?

Key Concepts Involved

Markov Transition Matrix (Stochastic Matrix): A square matrix \( P \) where each element \( P_{ij} \) represents the probability of transitioning from state \( i \) to state \( j \) in one step. Each row sums to 1.
Eigenvalue: A scalar \( \lambda \) such that for some nonzero vector \( v \), \( Pv = \lambda v \).
Stationary Distribution: A probability vector \( \pi \) such that \( \pi P = \pi \). This is an eigenvector with eigenvalue 1.

Mathematical Explanation

Let \( P \) be an \( n \times n \) Markov transition matrix. By definition, for each row:

\[ \sum_{j=1}^n P_{ij} = 1 \quad \forall i \]

Let \( \mathbf{1} \) be a column vector of all ones of length \( n \). Consider the action of \( P \) on \( \mathbf{1} \):

\[ P \mathbf{1} = \mathbf{1} \]

Thus, \( \mathbf{1} \) is a right eigenvector of \( P \) with eigenvalue 1.

Alternatively, if you consider the left eigenvector (the stationary distribution), a \( 1 \times n \) vector \( \pi \) such that:

\[ \pi P = \pi \]

This is also an eigenvector with eigenvalue 1.

Intuitive Explanation

In a Markov process, the total probability must be conserved. If you start with a probability distribution over states and advance it by one time step using the transition matrix, the total probability remains 1. This conservation is mathematically captured by the existence of eigenvalue 1.

Equation Summary

\[ P \mathbf{1} = \mathbf{1} \implies (P - I)\mathbf{1} = 0 \] \[ \text{So, 1 is always an eigenvalue of } P \]

Practical Implication

The eigenvalue 1 corresponds to the long-run, steady-state behavior of the Markov chain. The associated eigenvector (normalized to sum to 1) represents the stationary distribution—the probability distribution that remains unchanged by further application of the transition matrix.

3. If two assets become perfectly correlated, what happens to the covariance matrix?

Understanding the Question

You are asked to analyze the effect on the covariance matrix when two assets become perfectly correlated (their correlation coefficient becomes ±1).

Key Concepts Involved

Covariance Matrix: For \( n \) assets, a symmetric \( n \times n \) matrix where each entry \( \Sigma_{ij} \) is the covariance between asset \( i \) and asset \( j \).
Correlation: The correlation coefficient \( \rho_{ij} \) between assets \( i \) and \( j \) is \( \rho_{ij} = \frac{\Sigma_{ij}}{\sigma_i \sigma_j} \).
Perfect Correlation: \( \rho_{ij} = 1 \) (or \( -1 \)) means assets move exactly together (or in opposite directions) every period.
Linear Dependence: If two columns of a matrix are identical or negatives of each other, they are linearly dependent.
Matrix Rank: The rank of the covariance matrix decreases when two assets are perfectly correlated.

Mathematical Explanation

Suppose you have assets \( A \) and \( B \) with returns \( r_A \) and \( r_B \) and variances \( \sigma_A^2 \) and \( \sigma_B^2 \). If they are perfectly positively correlated (\( \rho_{AB} = 1 \)), then:

\[ r_B = a r_A + b \]

for some constants \( a \) and \( b \) (if the means are not zero).

The covariance between them is:

\[ \Sigma_{AB} = \rho_{AB} \sigma_A \sigma_B = \sigma_A \sigma_B \]

And for variances:

\[ \text{Var}(r_B) = a^2 \text{Var}(r_A) \]

Covariance Matrix Structure

The two columns (or rows) corresponding to \( A \) and \( B \) in the covariance matrix will be proportional:

\[ \begin{pmatrix} \sigma_A^2 & \sigma_A \sigma_B \\ \sigma_A \sigma_B & \sigma_B^2 \end{pmatrix} \] \quad \text{with } \sigma_A \sigma_B = \sqrt{\sigma_A^2 \sigma_B^2} \]

Thus, the second row is just a scaled version of the first row (and similarly for the columns). This means the two columns (rows) are linearly dependent.

Generalization to n Assets

Whenever two assets are perfectly correlated, their return vectors are perfectly colinear (one is a scalar multiple of the other).
The covariance matrix becomes singular (not invertible); its rank decreases by one for each pair of perfectly correlated assets.
In fact, if you have \( n \) assets, and \( k \) of them are all perfectly correlated with each other, these \( k \) assets contribute only one unique direction (dimension) to the covariance matrix.

Equation Summary

\[ \begin{align*} \text{If} \; r_B = a r_A + b \implies \text{Cov}(r_A, r_B) = a \text{Var}(r_A) \\ \text{Covariance matrix columns for } r_A, r_B \text{ are linearly dependent} \\ \implies \text{Rank decreases by 1} \end{align*} \]

Practical Implication

A singular covariance matrix cannot be inverted, which is problematic for portfolio optimization and risk calculations. In practice, perfect correlation rarely occurs, but near-perfect correlation can cause numerical instability and should be managed with regularization or by removing/reducing redundant assets.

Detailed Explanations and Insightful Examples

1. Factor Model Example: Visualizing Covariance Matrix Rank

Suppose you simulate returns for 100 stocks, each as a linear combination of 5 factors:


import numpy as np

n_stocks = 100
n_factors = 5
n_obs = 1000
Let’s simulate returns for 100 stocks driven by 5 independent factors, with no idiosyncratic noise, and examine the rank of the covariance matrix:



import numpy as np

n_stocks = 100
n_factors = 5
n_obs = 1000

# Simulate factor returns (n_obs x n_factors)
factor_returns = np.random.randn(n_obs, n_factors)

# Simulate random factor loadings for each stock (n_stocks x n_factors)
factor_loadings = np.random.randn(n_stocks, n_factors)

# Calculate stock returns (n_obs x n_stocks)
# Each stock's return is a linear combination of the factors
stock_returns = factor_returns @ factor_loadings.T

# Compute the sample covariance matrix (n_stocks x n_stocks)
cov_matrix = np.cov(stock_returns, rowvar=False)

# Calculate the rank of the covariance matrix
matrix_rank = np.linalg.matrix_rank(cov_matrix)

print(f"Rank of the covariance matrix: {matrix_rank}")



Expected output: The printed rank should be 5 (or very close, within numerical precision). This demonstrates that, in the absence of idiosyncratic risk, the covariance matrix is of rank equal to the number of independent factors.


2. Markov Transition Matrix: Eigenvalue 1 Example

Let’s create a simple Markov transition matrix and confirm the presence of eigenvalue 1:



import numpy as np

# Example: 3-state Markov chain with arbitrary transition probabilities
P = np.array([
    [0.5, 0.3, 0.2],
    [0.2, 0.6, 0.2],
    [0.3, 0.2, 0.5]
])

eigenvalues, eigenvectors = np.linalg.eig(P)
print("Eigenvalues of P:", eigenvalues)
# Check which eigenvalue is 1 (within numerical tolerance)
print("Eigenvalue 1 present:", np.any(np.isclose(eigenvalues, 1)))



Expected output: Among the eigenvalues printed, one will be exactly or extremely close to 1, confirming the theory.


Finding the Stationary Distribution

The stationary distribution \( \pi \) satisfies \( \pi P = \pi \). Numerically, we can find this as the left eigenvector associated with eigenvalue 1. In practice:


# The stationary distribution is the normalized left eigenvector for eigenvalue 1
stationary = eigenvectors[:, np.isclose(eigenvalues, 1)]
stationary = stationary[:, 0].real  # Take the real part if needed
stationary /= stationary.sum()      # Normalize so it sums to 1

print("Stationary distribution:", stationary)



This vector represents the long-run probabilities of being in each state.


3. Perfect Correlation: Covariance Matrix Example

Let’s construct a covariance matrix for two perfectly correlated assets and examine its properties:



import numpy as np

# Suppose asset A and asset B have standard deviations 2 and 3, respectively
sigma_A = 2
sigma_B = 3

# Perfect correlation: correlation coefficient = 1
corr_AB = 1.0

# Covariance = correlation * sigma_A * sigma_B
cov_AB = corr_AB * sigma_A * sigma_B

# Covariance matrix for A and B
cov_matrix = np.array([
    [sigma_A ** 2, cov_AB],
    [cov_AB, sigma_B ** 2]
])

print("Covariance matrix:\n", cov_matrix)

# Calculate rank
matrix_rank = np.linalg.matrix_rank(cov_matrix)
print("Rank of covariance matrix:", matrix_rank)



Expected output: The rank will be 1, not 2, because the two assets are perfectly correlated and thus linearly dependent.


Adding More Assets

If you have a portfolio of \( n \) assets, and two of them are perfectly correlated, the total rank of the covariance matrix falls by one. If all assets are perfectly correlated, the rank is 1.


Portfolio Optimization Implication

Most quantitative finance techniques, like mean-variance optimization, require inverting the covariance matrix. If the matrix is singular or nearly singular (due to perfect or near-perfect correlations), the inversion fails or becomes highly unstable. This is why regularization, shrinkage, or principal component analysis (PCA) is used in practice to address these issues.




Frequently Asked Questions (FAQs)

Q1. Why does the absence of idiosyncratic risk reduce the rank of the covariance matrix?

Idiosyncratic risk adds unique, independent variation to each asset’s return. Without it, all assets’ returns are just combinations of the same small set of factors, so the covariance matrix cannot have more independent directions than there are factors.


Q2. How do you find the number of independent factors from empirical data?

In practice, you can estimate the number of factors by performing eigenvalue decomposition (PCA) on the empirical covariance matrix. The number of significant (non-negligible) eigenvalues typically corresponds to the number of independent risk factors.


Q3. Can a Markov transition matrix have other eigenvalues equal to 1?

For an irreducible and aperiodic Markov chain, the eigenvalue 1 is unique. For reducible or periodic chains, you may observe multiple eigenvalues of 1.


Q4. What are the implications of perfect correlation for risk management?

Perfect correlation means no diversification benefit between the assets. In risk management, it signals redundancy: holding both assets is equivalent to holding one in scaled quantity.




Summary Table of Key Points



  
    
      Question
      Key Concept
      Detailed Explanation
      Practical Impact
    
  
  
    
      Covariance matrix of 100 stocks, 5 factors, no idiosyncratic risk
      Matrix rank limited by number of factors
      All returns are linear combinations of 5 factors; covariance matrix rank is 5
      Covariance matrix is singular; needs regularization for inversion
    
    
      Markov transition matrix eigenvalue 1
      Probability conservation, stationary distribution
      Rows sum to 1; eigenvalue 1 always exists
      Eigenvector for eigenvalue 1 is long-run state probabilities
    
    
      Perfect correlation between two assets
      Linear dependence, rank deficiency
      Covariance matrix columns/rows become proportional; rank drops by 1
      Inversion fails; signals redundant assets
    
  





Conclusion

A successful Graviton Quant Analyst candidate must not only know the answers to technical questions but also understand the theory and practical implications behind them. Whether dealing with factor models and covariance matrices, Markov processes and transition matrices, or the effects of asset correlation, a solid grasp of linear algebra, probability, and risk theory is essential. Mastering these concepts will give you a significant advantage in both the interview and your quantitative finance career.


Further Reading

  “Quantitative Investment Analysis” by DeFusco, McLeavey, Pinto, Runkle
  “Matrix Algebra Useful for Statistics” by Shayle Searle
  “Markov Chains” by J.R. Norris
  Research papers on covariance shrinkage and factor models in quantitative finance



For more practice, try generalizing these concepts to higher-dimensional problems, or implement your own factor model simulations and Markov processes in Python or R.



Related Articles
Quant Interview Questions from G-Research
Quant Interview Experience from Graviton
Matrix Questions from Quant Interviews

Question	Key Concept	Detailed Explanation	Practical Impact
Covariance matrix of 100 stocks, 5 factors, no idiosyncratic risk	Matrix rank limited by number of factors	All returns are linear combinations of 5 factors; covariance matrix rank is 5	Covariance matrix is singular; needs regularization for inversion
Markov transition matrix eigenvalue 1	Probability conservation, stationary distribution	Rows sum to 1; eigenvalue 1 always exists	Eigenvector for eigenvalue 1 is long-run state probabilities
Perfect correlation between two assets	Linear dependence, rank deficiency	Covariance matrix columns/rows become proportional; rank drops by 1	Inversion fails; signals redundant assets


                
                
                  
                  
                    
                  
                  
                  
                    
                    Data-Science
                    
                    Statistics
                    
                    Quant

Graviton Quant Analyst Interview: Key Questions and Answers

Graviton Quant Analyst Questions – Solved & Explained

1. You have returns of 100 stocks, but they are all generated by exactly 5 independent factors with no idiosyncratic component. What is the maximum rank of the covariance matrix?

Understanding the Question

Key Concepts Involved

Mathematical Explanation

What is the Rank of \( \Sigma_r \)?

Intuitive Explanation

Equation Summary

Practical Implication

2. A Markov transition matrix has eigenvalue 1. Why?

Understanding the Question

Key Concepts Involved

Mathematical Explanation

Intuitive Explanation

Equation Summary

Practical Implication

3. If two assets become perfectly correlated, what happens to the covariance matrix?

Understanding the Question

Key Concepts Involved

Mathematical Explanation

Covariance Matrix Structure

Generalization to n Assets

Equation Summary

Practical Implication

Detailed Explanations and Insightful Examples

1. Factor Model Example: Visualizing Covariance Matrix Rank

2. Markov Transition Matrix: Eigenvalue 1 Example

Finding the Stationary Distribution

3. Perfect Correlation: Covariance Matrix Example

Adding More Assets

Portfolio Optimization Implication

Frequently Asked Questions (FAQs)

Q1. Why does the absence of idiosyncratic risk reduce the rank of the covariance matrix?

Q2. How do you find the number of independent factors from empirical data?

Q3. Can a Markov transition matrix have other eigenvalues equal to 1?

Q4. What are the implications of perfect correlation for risk management?

Summary Table of Key Points

Conclusion

Further Reading

Related Articles

Adnan

Search

Recent Articles

Tags

Join Our Newsletter!

YouTube Channel