Portfolio Management interview preparation

Questions

100

Traced to a firm

40

Firms

24

Updated

September 2026

Asked at

All firmsBLBlackRock4Vanguard4WMWellington Management4Amundi3ACAQR Capital Management3Neuberger Berman3SCSchroders3Man Group2MSCI2Northern Trust2AllianceBernstein1Apollo Global Management1Blackstone1BMBNY Mellon1Carlyle Group1Fidelity Investments1Goldman Sachs1Invesco1Millennium Management1MSMorgan Stanley1NUNuveen1PIMCO1SSState Street1TPTPG1

Topic

All topicsPortfolio theory5Factor models8Asset allocation11Rebalancing3Portfolio construction7Benchmarks and tracking error5Performance measurement8Risk management6Fixed income and LDI5Currency and global3Implementation and costs5Active versus passive6India markets7Brainteasers5Career and fit16

Level

AnyCoreIntermediateHard

Type

AnyTechnicalCaseMarket viewFitBrainteaser

1. 028Walk me through mean-variance optimisation as you would actually run it.Portfolio constructionIntermediatetechnicalMulti-assetQuantitative asset management

   Say this

   You need three inputs, expected returns, volatilities and correlations, plus an objective and constraints. You maximise return for a given risk, or maximise the Sharpe ratio, and the output is a weight vector. In practice most of the work is fixing the inputs so the output is usable.

   Then walk it

   1. Formally: maximise w transpose mu minus lambda over two times w transpose sigma w, subject to weights summing to one and whatever bounds you impose. Lambda is risk aversion and tracing it out gives the frontier.
   2. Input one, expected returns, which is where almost all the error lives. I would build them from building blocks rather than history, and I would shrink them hard toward a common prior or use reverse optimisation from market weights.
   3. Input two, the covariance matrix. Sample covariance on 20 assets with 60 monthly observations is badly conditioned, so shrink it, Ledoit-Wolf being the standard, or impose a factor structure so you are estimating a few dozen numbers rather than 200.
   4. Then constraints, and be deliberate about them. Long-only, position caps, asset class ranges, turnover limits, liquidity limits. Constraints are how practitioners smuggle robustness into a fragile optimisation, and they generally help out of sample.
   5. Then diagnose the output before believing it. If a 0.2 percent change in one expected return moves the allocation by 15 percentage points, the answer is not a portfolio, it is an artefact. I would run resampling and report the distribution of optimal weights rather than the point solution.
   6. And the completeness check: compare the optimiser output against equal weight, against risk parity, and against the current portfolio. If it cannot beat those in a robustness test, use one of those instead. That is a normal outcome, not a failure.

   Where candidates lose it

   Reciting the maths and stopping at 'then you get the efficient frontier'. Everyone can do the formula. What distinguishes a usable answer is naming shrinkage of both inputs, constraints as a robustness device, and a sensitivity test on the output. Say which input the answer is most sensitive to before you are asked.

   Expect next

   • Which input dominates the error?
   • What is shrinkage doing, intuitively?
   • Why do constraints help out of sample?
2. 031What is risk parity, and what are its weaknesses?Portfolio constructionIntermediatetechnicalMulti-assetSystematic investing

   Say this

   Risk parity sizes positions so every asset contributes the same amount of risk, rather than the same amount of capital, and then levers the whole thing up to the required return. Its weaknesses are the leverage, the implicit assumption that risk-adjusted returns are equal, and a heavy structural exposure to bonds.

   Then walk it

   1. Construction: in the simple version, weights are inversely proportional to volatility. In the proper version you equalise marginal contribution to risk, which accounts for correlations, so a cluster of correlated assets gets less weight in total.
   2. The motivation is that 60/40 is not a balanced portfolio at all. Equities at 15 percent volatility and bonds at 5 percent means roughly 90 percent of the risk sits in the equity sleeve, so the portfolio is an equity portfolio with a modest cushion.
   3. Because low-volatility assets get large weights, the unlevered portfolio has too little expected return, so leverage is applied, usually through futures and repo. The claimed payoff is a higher Sharpe ratio at the same volatility, which is the same betting-against-beta logic as low-vol equity investing.
   4. Weakness one, the return assumption. Equal risk contribution is only optimal if Sharpe ratios are equal and correlations are similar. That is an assumption, just a less visible one than in mean-variance.
   5. Weakness two, leverage and funding. You now have financing cost, margin calls, and the possibility of forced deleveraging into a falling market. Risk parity funds had a bad 2022 for exactly this reason, because bonds and equities fell together and the levered bond leg amplified it.
   6. Weakness three, the bond dependence. Forty years of falling yields flattered the strategy. At low or negative real yields the bond leg has poor expected return and asymmetric risk, so the historical Sharpe is not a forward-looking estimate. Volatility targeting also makes it procyclical, adding risk in calm markets and cutting it after volatility spikes, which mechanically sells the bottom.

   Where candidates lose it

   Describing it as 'equal risk from each asset' and stopping, or presenting it as assumption-free. It has strong embedded assumptions, equal Sharpe ratios, and it relies on leverage and on the bond leg behaving. If you cannot explain why 2022 was bad for risk parity, the answer sounds theoretical.

   Expect next

   • Why did risk parity struggle in 2022?
   • How much leverage would it need for 10 percent volatility?
   • How is risk parity related to betting against beta?
3. 032How do constraints like no shorting and maximum position size change the optimisation?Portfolio constructionIntermediatetechnicalQuantitative asset managementPortfolio implementation

   Say this

   They lower the theoretical efficient frontier and they usually raise realised performance. In theory constraints can only cost you, since the unconstrained solution is feasible in a larger set. In practice they protect you from estimation error, which is the bigger enemy.

   Then walk it

   1. The theory: every binding constraint shifts the frontier down and to the right. A long-only constraint is particularly expensive for a signal whose information is concentrated in the short leg, because you can only express a negative view by going to zero weight.
   2. The practical effect runs the other way. Constraints stop the optimiser from acting on its most extreme, least reliable estimates, so out of sample constrained portfolios routinely beat unconstrained ones. Jagannathan and Ma showed the long-only constraint acts like a form of shrinkage on the covariance matrix.
   3. The cost is not symmetric across the universe. For a large cap portfolio a long-only constraint costs little, because index weights give you room to underweight. For small caps, where a name might be 0.05 percent of the index, the maximum possible underweight is trivial, so the constraint binds hard.
   4. Constraints also change what the transfer coefficient is. The fundamental law with a transfer coefficient says realised information ratio equals the coefficient times skill times the square root of breadth, and a constrained portfolio typically has a transfer coefficient of 0.3 to 0.6, so you are capturing half your signal at best.
   5. So I would treat constraints as a portfolio decision rather than a compliance afterthought, and quantify the cost: run the optimisation with and without each constraint and see what it costs in expected information ratio. Then you can argue for relaxing the expensive ones, for example allowing 130/30 or a modest net short capability.
   6. And say the governance reality: many constraints exist because a client or a regulator requires them, so the job is to build the best portfolio inside them and to be able to price what they cost.

   Where candidates lose it

   Saying constraints are simply bad because they reduce the opportunity set. That is only true if your inputs are correct, which they are not. The sophisticated answer names the shrinkage effect and quantifies the loss through the transfer coefficient, and then says which constraints are worth arguing about.

   Expect next

   • What is a transfer coefficient?
   • Where does a long-only constraint cost you most?
   • How would you price the cost of a constraint to a client?
4. 034How does ESG integration actually change portfolio construction?Portfolio constructionIntermediatetechnicalAsset managementSustainable investing

   Say this

   It adds a constraint or a tilt to the optimisation, and the important question is which. Exclusion is a hard constraint that costs tracking error and creates unintended sector and factor bets. Integration treats ESG data as another input to expected returns and risk, which changes weights without necessarily removing names.

   Then walk it

   1. Distinguish the approaches, because the interviewer is testing whether you can: exclusion, best-in-class tilting, integration into the fundamental view, thematic allocation, and engagement plus voting. They have completely different portfolio consequences.
   2. Exclusion is the expensive one. Removing energy and tobacco from a global benchmark does not just remove those names, it creates a growth tilt, a duration tilt and a value underweight. So I would run an optimiser that maximises the ESG score improvement per unit of tracking error, rather than just deleting names.
   3. Integration is cheaper and more defensible analytically. Carbon transition risk becomes a cash flow and terminal value assumption; governance quality becomes part of the discount rate or a haircut to reported earnings. That changes what you pay, not whether you own it.
   4. Data quality is the real constraint. ESG ratings from two providers correlate around 0.5, far below the 0.99 you get between credit rating agencies, so a portfolio built to one provider's score is partly built to that provider's methodology. I would use raw underlying metrics, emissions intensity, board independence, accident rates, rather than composite scores.
   5. Then measure the cost honestly. Report the tracking error the ESG constraint introduces and the factor tilts it creates, so the client can decide whether the objective is worth the active risk. A 1 percent tracking error to cut portfolio carbon intensity by half is a defensible trade; hiding it is not.
   6. And on returns: the fair position is that ESG is neither a reliable alpha source nor a guaranteed cost. Governance quality has decent evidence as a risk factor, exclusion reduces the opportunity set, and much of the historical outperformance of ESG funds is explained by a growth and quality tilt rather than by the ESG data itself.

   Where candidates lose it

   Treating ESG as either a marketing exercise or a guaranteed performance edge. The credible answer is technical: which approach, what it costs in tracking error, what unintended factor bets it creates, and the fact that provider ratings disagree. Naming the low correlation between ESG rating providers is the detail that shows you have handled the data.

   Expect next

   • How would you build a low-carbon version of an index with minimal tracking error?
   • Does ESG cost you return?
   • How do you handle two providers disagreeing on the same company?

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.