← Morgan Stanley Interview Insights
Start by anchoring the option's moneyness (ATM, ITM, OTM) using the spot-strike relationship, then apply well-known approximations for each Greek rather than trying to derive Black-Scholes from scratch. Communicate your reasoning transparently, stating assumptions and acceptable error ranges, which demonstrates both quantitative intuition and engineering pragmatism.
Pro tip: Interviewers at quant-heavy firms like Morgan Stanley care more about your structured reasoning and sanity-checking instincts than arithmetic precision — explicitly call out the ATM approximations (delta ≈ 0.5, gamma and vega are maximized) and explain why, which signals you understand the Greeks geometrically, not just formulaically.
Compute the rough moneyness ratio S/K and determine if the option is ATM (S≈K), ITM, or OTM. This single classification drives all subsequent approximations, since Greeks behave very differently across these regimes.
For a call, use delta ≈ 0.5 if ATM, trending toward 1.0 deep ITM and 0.0 deep OTM; for a put, mirror with negative sign. Refine by mentally computing d1 = [ln(S/K) + (r + σ²/2)T] / (σ√T) and mapping it to the standard normal CDF using the rule that N(0)=0.5, N(1)≈0.84, N(-1)≈0.16.
Use the ATM approximation gamma ≈ N'(d1) / (S·σ·√T), where N'(0) ≈ 0.4 (the standard normal PDF peak). Gamma is highest ATM and decays for ITM/OTM options, and also shrinks as time-to-expiry or volatility increases.
Apply vega ≈ S·√T·N'(d1), which for an ATM option simplifies to roughly S·√T·0.4. Note that vega is also maximized ATM, is symmetric for calls and puts, and scales with the square root of time — a key intuition to verbalize.
Verify internal consistency: delta should be between 0 and 1, gamma and vega should be positive, and put-call parity implies call delta minus put delta equals 1. State your confidence interval and identify which approximations introduce the most error (e.g., ignoring higher-order terms in d1).
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.
Start by anchoring your answer with the key landmark values of the standard normal CDF and PDF that any quant-oriented engineer should have memorized, then explain practical interpolation strategies between those anchors. Demonstrate awareness of the trade-off between accuracy and computational simplicity, and mention polynomial or rational approximations used in production systems.
Pro tip: Mentioning the 68-95-99.7 rule is expected, but what impresses interviewers at firms like Morgan Stanley is knowing Φ(1.645)≈0.95 and Φ(2.326)≈0.99 — the one-tailed VaR cutoffs used daily in risk management — showing you understand why these approximations matter in finance.
Recite the key CDF values you know by heart: Φ(0)=0.5, Φ(1)≈0.8413, Φ(1.645)≈0.95, Φ(1.96)≈0.975, Φ(2)≈0.9772, Φ(2.576)≈0.995, Φ(3)≈0.9987. For the PDF, note φ(0)≈0.3989 and that it falls off as a Gaussian with σ=1.
Explain that for values between known anchors, linear interpolation gives a reasonable first approximation — e.g., Φ(1.5) ≈ 0.5*(Φ(1)+Φ(2)) ≈ 0.9093, which is close to the true 0.9332. Acknowledge this introduces error in the tails where the CDF is more curved.
Mention that the CDF has no closed form but can be expressed via the error function: Φ(x) = 0.5*(1 + erf(x/√2)). Abramowitz & Stegun provide rational polynomial approximations (e.g., Hart's approximation) accurate to 7+ decimal places, which are what numerical libraries actually implement.
For large |x|, use the asymptotic Mills' ratio approximation: 1-Φ(x) ≈ φ(x)/x for x>>0, which is critical for tail-risk calculations. Note that linear interpolation breaks down in the tails and a log-linear or asymptotic approach is more appropriate.
Summarize when each method is appropriate: memorized values for quick sanity checks, linear interpolation for moderate precision in the body of the distribution, and polynomial approximations for production code requiring accuracy and speed without table lookups.
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.
This is the part I actually felt okay about.
Start by briefly recapping the Black-Scholes intuition for each Greek, then systematically walk through how each one changes as you perturb spot, vol, and time-to-expiry one at a time. Use concrete numerical examples or order-of-magnitude estimates to demonstrate sanity-checking discipline, which is exactly what a quant-adjacent SWE at Morgan Stanley needs to show.
Pro tip: Interviewers love when candidates catch their own mistakes by cross-checking Greeks against each other — for example, noting that a very high gamma near expiry should imply a rapidly changing delta, and verifying that the two estimates are mutually consistent. This shows you think in systems, not in isolated formulas.
Briefly state what delta (∂V/∂S), gamma (∂²V/∂S²), and vega (∂V/∂σ) measure in plain terms. Anchor the audience before diving into scaling behavior so the rest of your answer has a clear foundation.
Explain that delta for a call moves from ~0 (deep OTM) to ~1 (deep ITM) as S rises, gamma peaks near ATM and falls off on both sides, and vega is also maximized ATM and decreases as the option moves deep ITM or OTM. Use the intuition that moneyness (S/K) is the key driver.
Higher vol 'spreads' the distribution, pushing delta of OTM options toward 0.5 and reducing gamma (since the peak flattens), while vega itself increases for near-ATM options but has a more complex shape. Note that vega ∝ S√T in the BS formula, giving a concrete scaling handle.
As T decreases, gamma and vega both spike for ATM options (gamma ∝ 1/(S·σ·√T)) while delta becomes a step function at expiry. Conversely, long-dated options have smoother, smaller gamma and larger vega in absolute terms due to the √T factor.
Pick a concrete example (e.g., ATM call, S=100, σ=20%, T=1yr) and perturb one parameter at a time — bump S by 1%, double σ, halve T — then verify the direction and rough magnitude of each Greek's change against your scaling rules. Flag any inconsistency as a red flag that would prompt deeper investigation in a real system.
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.