Risk Management interview preparation

Questions

100

Traced to a firm

37

Firms

12

Updated

September 2026

Asked at

All firmsUBS14MSCI7BLBlackRock5FTFranklin Templeton3Oaktree Capital Management2Scotiabank2Jane Street1Moody's1Neuberger Berman1PIMCO1SSState Street1TSTruist Securities1

Topic

All topicsMarket risk and VaR14Tail risk and stress testing5Greeks and sensitivities5Credit risk11Counterparty risk and CVA6Operational risk5Model risk and validation6Regulatory capital7Liquidity risk and ALM6Statistics and quant foundations7Indian regulation7Risk governance and appetite4Markets and macro9Fit and career8

Level

AnyCoreIntermediateHard

Type

AnyTechnicalCaseBrainteaserMarket viewFit

1. 006How does Monte Carlo VaR work, and when is it worth the extra cost?Market risk and VaRHardtechnicalUBSRisk Management · Zurich · 2021

   Say this

   You specify a stochastic process for each risk factor, simulate a large number of joint paths, revalue the portfolio on every path, and take the percentile of the simulated P&L. It's worth the cost when the payoff is non-linear or path dependent, and not otherwise.

   Then walk it

   1. The inputs are a process per factor, usually a drift and volatility, plus a dependence structure, usually a correlation matrix or a copula. Then you draw correlated shocks, typically via a Cholesky decomposition.
   2. Full revaluation is the expensive part, not the random numbers. If revaluing one exotic takes a second, ten thousand paths across a thousand trades is a real overnight compute problem.
   3. It's the only method that handles path dependency properly. A barrier option, a cliquet, a callable bond, a CVA number on a swap portfolio, all of those depend on the path and not just the endpoint.
   4. It also lets you choose the distribution. You can simulate from a t distribution, or use a copula to get tail dependence that a normal correlation matrix cannot produce.
   5. Its weakness is that it is only as good as the assumed process. Historical simulation is wrong in a way you can see; Monte Carlo is wrong in a way buried in a calibration file. That's why it needs the heaviest model validation of the three.
   6. So my rule: linear portfolio, don't bother, parametric or historical is fine. Options book, structured credit, or anything with optionality in the funding, Monte Carlo earns its keep.

   Where candidates lose it

   Describing it as 'generating random scenarios' without naming the two things you have to assume, the process and the dependence structure. That's where all the model risk lives, and naming it is what separates someone who has built one from someone who read about it.

   Expect next

   • How many paths do you need, and how would you know?
   • How would you introduce tail dependence into the simulation?
   • How would you validate a Monte Carlo VaR engine?

   Reported by candidates at UBS (Risk Management, Zurich, 2021). Source: Wall Street Oasis.

2. 007Which assumption inside parametric VaR fails first, and what does that do to your number?Market risk and VaRHardsuperdayUBSRisk Management · Zurich · 2021

   Say this

   Normality fails first, and it makes VaR too small exactly when you need it. Real return distributions are leptokurtic, so the true 99th percentile sits further out than 2.33 sigma, and the deeper into the tail you go the worse the understatement gets.

   Then walk it

   1. Assumption one, normality. Equity index daily returns have kurtosis well above three. At 99% the error is modest, maybe 10 to 20 percent; at 99.9% parametric VaR can be off by a factor.
   2. Assumption two, a stable covariance matrix. Correlations rise in a sell-off, so the diversification benefit the matrix gives you evaporates in the scenario the number is supposed to protect you from.
   3. Assumption three, linearity. Parametric VaR uses deltas, so it prices an option position as if it were stock. Short gamma looks harmless and short a straddle can even show negative risk.
   4. Assumption four, independent returns. Volatility clusters, so square-root-of-time scaling understates multi-day risk during a stress period.
   5. The order matters for the answer: normality is the one people name, but linearity is the one that produces catastrophically wrong numbers, because it can be wrong by a sign rather than a percentage.
   6. Fixes in ascending order of effort: a t distribution or Cornish-Fisher adjustment for the tail, EWMA covariance for the clustering, delta-gamma for mild convexity, and full revaluation once the book has real optionality.

   Where candidates lose it

   Saying 'it assumes normality' and stopping. Every candidate says that. The differentiator is naming the linearity assumption and explaining that for an options book parametric VaR can get the direction of risk wrong, not just the magnitude.

   Expect next

   • How would you adjust it for fat tails without going to full simulation?
   • What does delta-gamma VaR fix and what does it still miss?
   • Would you ever show a board a parametric number? When?

   Reported by candidates at UBS (Risk Management, Zurich, 2021). Source: Wall Street Oasis.

3. 009What are the advantages and disadvantages of expected shortfall compared with VaR?Market risk and VaRHardtechnicalUBSRisk Management · Zurich · 2021

   Say this

   ES wins on theory and loses on practice. It's coherent, it sees the whole tail, and it can't be gamed by moving risk past the threshold. But it's harder to backtest, less stable, and more sensitive to the handful of observations that drive it.

   Then walk it

   1. Advantage one, it's subadditive, so it's a coherent risk measure. Adding two books can never raise ES above the sum of their parts, which means you can allocate it down to desks and the numbers add up sensibly.
   2. Advantage two, it sees tail depth. VaR is blind beyond the quantile, so a desk can sell far out-of-the-money options and report the same VaR with vastly more real exposure. ES prices that in.
   3. Advantage three, it removes the incentive to optimise against the measure. Optimising a portfolio to minimise VaR tends to push loss into the tail; minimising ES doesn't reward that.
   4. Disadvantage one, backtesting. A VaR breach is binary and you can test the count with a Kupiec or traffic-light test. ES needs you to test conditional magnitudes, which needs far more observations, so supervisors still backtest VaR even under an ES capital regime.
   5. Disadvantage two, estimator noise. At 97.5% over 250 days, ES is the average of six observations. Change one bad day and the number jumps. It's less robust and less stable period to period, which makes limit management awkward.
   6. Disadvantage three, communication. Traders understand 'I lose more than this one day in a hundred'. 'The average of my worst six days' takes longer to land, and risk numbers nobody understands don't change behaviour.
   7. My summary line: ES is the better measure of risk and VaR is the better test of your model. Most banks now report both for exactly that reason.

   Where candidates lose it

   Giving only the coherence advantage. That's half the answer and the easy half. The interviewer is testing whether you know the practical cost, and the backtesting problem is the answer. Saying 'ES is strictly better' is the wrong answer, because if it were, Basel would have dropped VaR backtests too.

   Expect next

   • If ES is coherent and VaR is not, why do supervisors still backtest VaR?
   • How many observations would you want to estimate ES reliably?
   • Which would you set a desk limit on?

   Reported by candidates at UBS (Risk Management, Zurich, 2021). Source: Wall Street Oasis.

4. 010Why is VaR not a coherent risk measure?Market risk and VaRHardsuperdayUBSRisk Management · Zurich · 2021

   Say this

   Because it fails subadditivity. The VaR of a combined portfolio can exceed the sum of the individual VaRs, which means diversification can appear to increase risk. Coherence needs four properties, and that's the one VaR breaks.

   Then walk it

   1. The four axioms are monotonicity, translation invariance, positive homogeneity and subadditivity. VaR satisfies the first three.
   2. The classic counterexample is two independent digital or deep out-of-the-money option positions. Each has a small probability of a large loss, say 0.6 percent. Individually, at 99% confidence, the loss sits beyond the quantile, so each has near-zero VaR.
   3. Put them together and the probability of at least one blowing up is now above one percent, so the combined VaR jumps to the full loss. Two positions with almost no VaR each combine into a large one. That's the violation.
   4. Why it matters operationally: if the measure isn't subadditive, you can't safely allocate a firm limit down to desks, because desk limits summing to the firm limit no longer bound the firm's risk. And a trader can reduce measured VaR by taking on tail risk.
   5. For elliptical distributions, including the normal, VaR is subadditive, which is why the problem never shows up in a textbook example. It shows up in real books with credit and optionality, which is exactly where it matters.
   6. ES is subadditive at every confidence level and for every distribution, which is the theoretical reason Basel moved to it.

   Where candidates lose it

   Naming subadditivity without being able to construct the counterexample. The interviewer will ask for an example, and 'two out-of-the-money digital options that each blow up 0.6 percent of the time' is the one that works. Also worth avoiding: claiming VaR is never subadditive. For normal distributions it is.

   Expect next

   • Give me a concrete two-position counterexample.
   • Is VaR subadditive under any conditions?
   • What practical problem does this create for limit setting?

   Reported by candidates at UBS (Risk Management, Zurich, 2021). Source: Wall Street Oasis.

5. 050Explain what a Kalman filter is.Model risk and validationHardtechnicalUBSRisk · London · 2022

   Say this

   It's a recursive estimator for a hidden state you can only observe with noise. Each period you predict the state forward with your model, then correct that prediction with the new observation, weighting the two by how much you trust each. Under linear-Gaussian assumptions it's the optimal estimator.

   Then walk it

   1. Two equations. A state equation for how the unobserved thing evolves, and a measurement equation linking the state to what you actually see, each with its own noise.
   2. Two steps per period. Predict: roll the state and its uncertainty forward. Update: compute the surprise, the difference between the observation and what you expected, and move your estimate toward it by the Kalman gain.
   3. The gain is the whole intuition. If measurement noise is large relative to state uncertainty, the gain is small and you mostly trust your model. If your state uncertainty is large, the gain is large and you mostly trust the new data. It's Bayesian updating with the arithmetic done for you.
   4. Where it's used in finance: extracting a time-varying beta or hedge ratio, estimating a stochastic volatility or unobserved factor, filtering a fair-value or pairs-trading spread, term structure models where the factors are latent, and nowcasting a macro variable from noisy high-frequency data.
   5. Why a risk function cares: it gives you an estimate that adapts without the jumpiness of a rolling window. A 60-day rolling beta lurches when an old observation drops out; a Kalman-filtered beta moves smoothly and quantifies its own uncertainty.
   6. The assumptions and their cost: linear dynamics and Gaussian noise. For non-linear problems you need the extended or unscented variants or a particle filter. And you have to specify the two noise covariances, which are rarely known, so in practice you estimate them by maximum likelihood and the result is sensitive to them.
   7. The limitation to volunteer: it's optimal given the model, and it has no way to tell you the state equation is wrong. Feed it a misspecified process and it will produce confident, smooth, wrong estimates, which is a particularly dangerous failure mode.

   Where candidates lose it

   Reciting matrix equations. Nobody wants the algebra; they want the predict-then-correct intuition, the gain as a trust weighting, and one concrete financial use. If you can't name a use case, the answer reads as memorised from a signal-processing course.

   Expect next

   • How would you use it to estimate a time-varying hedge ratio?
   • What happens if the noise covariances are misspecified?
   • How does it compare to a simple exponentially weighted estimate?

   Reported by candidates at UBS (Risk, London, 2022). Source: Wall Street Oasis.

6. 066What does it mean if an estimator is BLUE?Statistics and quant foundationsHardtechnicalUBSRisk Management · Zurich · 2021

   Say this

   Best Linear Unbiased Estimator. Among all estimators that are linear in the data and unbiased, it has the smallest variance. That's the Gauss-Markov result: ordinary least squares is BLUE provided a specific set of assumptions holds.

   Then walk it

   1. Unpack each word, because that's what the question is testing. Linear in the observations. Unbiased, so its expected value equals the true parameter. Best, meaning minimum variance within that class.
   2. The Gauss-Markov conditions: correct linear specification, errors with zero conditional mean, homoskedasticity, no autocorrelation, and no perfect multicollinearity. Notice normality is not required for BLUE. You need normality for the t and F tests in small samples, not for OLS to be efficient.
   3. The restriction that matters is 'linear'. A biased or non-linear estimator can easily beat OLS on mean squared error. Ridge and lasso are deliberately biased and often predict better, and James-Stein shrinkage famously dominates the sample mean. So BLUE is optimality within a box, not optimality.
   4. In finance the assumptions fail routinely. Returns are heteroskedastic and volatility clusters, so OLS stays unbiased but the standard errors are wrong, which means your t-statistics lie. That's the practical consequence and it's the one to lead with when asked what breaks.
   5. The fixes: White or Newey-West robust standard errors for heteroskedasticity and autocorrelation, generalised least squares if you know the error structure, and instrumental variables if the regressor is endogenous. Endogeneity is the serious one, because it destroys unbiasedness rather than just efficiency.
   6. The distinction to keep straight: heteroskedasticity and autocorrelation cost you efficiency and valid inference. Omitted variables, measurement error in a regressor and simultaneity cost you unbiasedness. Those are different problems needing different fixes.
   7. So the answer I'd close with: BLUE is a useful benchmark and a weak guarantee. In a risk model I care more about whether the specification is right and whether the relationship is stable than about being efficient within the linear unbiased class.

   Where candidates lose it

   Expanding the acronym and stopping, or claiming normality is a Gauss-Markov requirement. It isn't. The two answers that separate candidates are that 'best' is only within linear unbiased estimators, so biased shrinkage estimators can beat it, and that in finance the binding violation is heteroskedasticity making your standard errors wrong.

   Expect next

   • Is normality required for OLS to be BLUE?
   • Which Gauss-Markov assumption fails most often in financial data?
   • Can a biased estimator ever be preferable?

   Reported by candidates at UBS (Risk Management, Zurich, 2021). Source: Wall Street Oasis.

Firm tags come from public, anonymous candidate reports on Wall Street Oasis: strong signal, not sworn testimony. Firms are named as the places a question was reported, not as partners of Fin Maverick. Answers are written for this page to show how to think out loud; they are not scripts to recite.