Capital One Assessment Test Advanced Applications 4 — Questions and Answers
Question 1: A Capital One product team is analyzing customer lifetime value (CLV). If average monthly revenue per customer is $45, monthly churn rate is 3%, and gross margin is 70%, what is the CLV?
- $1,050 (Correct answer)
- $900
- $1,500
- $315
Correct answer: $1,050
CLV = (Monthly revenue × Gross margin) / Churn rate = ($45 × 0.70) / 0.03 = $31.50 / 0.03 = $1,050.
Question 2: Which Python pandas method efficiently fills missing values in a time-series transaction dataset using the previous valid observation?
- df.fillna(method='ffill') (Correct answer)
- df.fillna(df.mean())
- df.dropna()
- df.interpolate(method='linear')
Correct answer: df.fillna(method='ffill')
ffill (forward fill) propagates the last valid observation forward, which preserves the last known state in a time series.
Question 3: An analyst is evaluating two credit models. Model A has AUC-ROC of 0.82; Model B has AUC-ROC of 0.79 but captures 15% more high-risk customers. Which factor should drive the final model selection?
- Business objective — if minimizing defaults is the priority, higher recall on high-risk customers may outweigh AUC (Correct answer)
- Always select the higher AUC model regardless of business context
- Always select the model with higher recall regardless of AUC
- Flip a coin since the difference is within statistical noise
Correct answer: Business objective — if minimizing defaults is the priority, higher recall on high-risk customers may outweigh AUC
Model selection depends on business objectives; a lower AUC model that catches more high-risk customers may reduce losses more than one with marginally higher AUC.
Question 4: Capital One's data warehouse uses a star schema. In this context, what is the role of a fact table?
- Stores quantitative business metrics (transactions, amounts, counts) with foreign keys to dimension tables (Correct answer)
- Stores descriptive attributes about business entities like customers and products
- Serves as the primary key source for all other tables
- Contains audit logs of ETL pipeline runs
Correct answer: Stores quantitative business metrics (transactions, amounts, counts) with foreign keys to dimension tables
Fact tables hold measurable business events and metrics with foreign keys linking to dimension tables that describe those events.
Question 5: A logistic regression model for credit approval returns a probability of 0.73 for a customer. With a decision threshold of 0.60, what is the model's prediction?
- Approve — 0.73 exceeds the 0.60 threshold (Correct answer)
- Deny — 0.73 is too close to the threshold to be reliable
- Approve — only if probability exceeds 0.80
- Deny — probability must equal 1.0 for approval
Correct answer: Approve — 0.73 exceeds the 0.60 threshold
When the predicted probability exceeds the decision threshold (0.73 > 0.60), the model classifies the outcome as the positive class (approve).
Question 6: Which SQL clause correctly calculates the percentage of total spend each category represents in a single query without a subquery?
- SUM(amount) / SUM(amount) OVER () * 100 AS pct_of_total (Correct answer)
- SUM(amount) / COUNT(*) * 100 AS pct_of_total
- SUM(amount) * 100 / MAX(SUM(amount)) AS pct_of_total
- amount / SUM(amount) GROUP BY category * 100 AS pct_of_total
Correct answer: SUM(amount) / SUM(amount) OVER () * 100 AS pct_of_total
A window function SUM() OVER () calculates the grand total in the same query, enabling single-pass percentage computation.
Question 7: In Capital One's risk management, a Monte Carlo simulation is used to model credit loss distributions. What is the primary advantage of this technique over a single-point estimate?
- It quantifies the full range of possible outcomes and their probabilities, capturing tail risk (Correct answer)
- It guarantees the most accurate single prediction of future losses
- It eliminates model uncertainty by running thousands of identical calculations
- It reduces computation time compared to analytical models
Correct answer: It quantifies the full range of possible outcomes and their probabilities, capturing tail risk
Monte Carlo simulation samples from probability distributions thousands of times to produce a loss distribution, revealing tail risk that single-point estimates miss.
A Capital One product team is analyzing customer lifetime value (CLV).
If average monthly revenue per customer is $45, monthly churn rate is 3%, and gross margin is 70%, what is the CLV?