Model Risk: VaR & Monte Carlo Orals

Model Risk Validation Manager loop — niche finance / risk-theory round. Recurring questions: VaR for long vs short positions, Monte Carlo path-count convergence. The interviewer probes one layer deeper on each surface answer; precision in terminology is the graded signal. Requirements The round spans model-validation orals and applied risk reasoning: VaR for long vs short positions. "If you're long $1M in a stock with VaR = $100K, what's the VaR if you short the same amount?" The answer is not simply $100K. Discuss: Return distributions are asymmetric — stocks drift upward historically, so the short loss tail differs from the long loss tail in magnitude and frequency. Short positions carry margin-call risk; LTV breaches force additional collateral, increasing effective exposure beyond the initial position. Always simulate returns, not prices — returns are stationary; prices are not. Short VaR is typically larger than long VaR when liquidity and margin risk are properly accounted for. Monte Carlo path count. "How many paths do you need?" Frame it as a convergence problem, not a fixed number: Stability of quantile estimates improves with √N — diminishing returns. At 95% confidence with 200K paths → 10K tail observations → stable estimate. Distinguish CLT / t-stats for parameter estimation (mean and σ from historical returns) from Monte Carlo path count for convergence of the quantile estimate. Validate by plotting VaR estimates across increasing path counts; stop when the change is within tolerance. Model-validation methodology. Be ready to discuss backtesting frameworks, traffic-light tests, model performance under stress scenarios, the Capital One / Discover merger context (additional model-validation complexity from combining institutions). Notes The round is at a level that expects production model-validation experience; this is a senior / specialist loop, not an entry-level role. Precision in terminology matters: know the difference between parameter estimation and simulation convergence; synthetic control method vs geo test; expected shortfall vs VaR; historical simulation vs parametric vs Monte Carlo VaR. For the VaR long-vs-short question, name the asymmetry up front — most candidates give the textbook $100K answer and lose the round in the first 30 seconds. Mention the regulatory context (Basel III, Fed CCAR / DFAST) when relevant; Capital One is regulated and the model-risk function reports up through the regulatory chain. Preparation Refresh the VaR / ES / parametric-vs-historical / Monte Carlo trade-offs; know the formulas and assumptions for each. Drill the convergence-of-quantile-estimates argument; this is the second-most-common probe in the round. Read the Federal Reserve's SR 11-7 model risk management guidance once before the loop; it is the regulatory anchor every Capital One model-risk conversation references implicitly.