Portfolio Management interview preparation
Asset allocation, factor models, risk, attribution and implementation, on global and Indian portfolios. Every question is either traced to a named firm from a public candidate report, or tagged at desk level when we could not trace it, and answers lead with the point, then the mechanism, then the limitation.
100 questions, mapped to the firms that asked them
- Questions
- 100
- Traced to a firm
- 40
- Firms
- 24
- Updated
- September 2026
010Explain how you would construct a factor, and then how you would optimise that construction.AQR Capital ManagementInvestment Research · New York · 2021
Say this
Build the raw signal first, rank the universe, neutralise the exposures you do not want, then form the long-short portfolio. Optimising it is mostly about the same premium surviving costs: slower rebalancing, buffer zones, better weighting and trading the cheap end of the signal.
Then walk it
- Start with signal hygiene: point-in-time data with no look-ahead, lagged fundamentals by the real reporting delay, winsorise outliers, and standardise to z-scores within the universe.
- Neutralise deliberately. Sector, country, size and beta neutrality strip out exposures you are not trying to own. A naive value screen in India is mostly a bet on PSUs and metals.
- Then the weighting. Quintile sorts are the academic default but they throw away information. A signal-weighted or optimised portfolio captures more of the spread for the same turnover, so long as you cap single-name weights.
- Optimisation is largely cost optimisation. Buffer the rebalance so names near the boundary do not churn, trade over multiple days, net the new signal against the existing book rather than liquidating, and rebalance on a schedule that matches the signal's decay. Momentum decays in weeks so it must be traded fast; value decays over years so daily rebalancing is pure cost.
- Then combine signals before trading, not after. Running separate value and momentum sleeves means you buy and sell the same name twice. Integrating the scores into one target portfolio cuts turnover materially, often by a third.
- And measure the optimisation honestly. Every knob you turn is another degree of freedom, so I would judge the final construction on out-of-sample and out-of-region performance net of realistic costs, not on the backtest Sharpe.
Where candidates lose it
Describing quintile sorts as if that were the whole job. At a firm like this the interesting content is implementation: turnover, buffering, signal integration and cost. Also, do not forget point-in-time data. A candidate who backtests on restated fundamentals has produced a number that cannot be traded.
Expect next
- How much would costs eat from a monthly rebalanced momentum factor?
- How do you decide the rebalancing frequency?
- How do you avoid overfitting when you have tuned this many parameters?
Reported by candidates at AQR Capital Management (Investment Research, New York, 2021). Source: Wall Street Oasis.
012You regress a fund's returns on factors and the OLS assumptions are violated. How would you fix it?AQR Capital ManagementInvestments · Greenwich · 2022
Say this
Diagnose which assumption broke, because the fix differs. Heteroskedasticity and autocorrelation do not bias the coefficients, only the standard errors, so you correct the errors. Omitted variables and endogeneity bias the coefficients themselves, and that needs a change of specification.
Then walk it
- Heteroskedasticity, which is guaranteed in financial returns because volatility clusters: keep OLS coefficients and use White or Newey-West robust standard errors. Or model the variance directly with GARCH if you care about the conditional risk.
- Autocorrelated residuals, common when the fund holds illiquid or stale-priced assets: Newey-West with a sensible lag, and separately run the Dimson or Scholes-Williams correction with lagged market returns, because stale marks understate beta. Summing the lagged betas is often the real finding.
- Multicollinearity between factors, for example value and profitability after 2015: coefficients stay unbiased but become unstable and standard errors blow up. Orthogonalise the factors, drop one, or use ridge rather than pretending the loadings are precise.
- Omitted variable, which is the dangerous one: a fund with apparent alpha against a three factor model often loses all of it against a five factor model with momentum. That is bias, not inefficiency, so the fix is a better specification, not better standard errors.
- Non-normal residuals and fat tails: OLS is still consistent, but inference on short samples is unreliable, so I would bootstrap the confidence intervals rather than trusting t-statistics from 36 monthly observations.
- And the structural one nobody tests for: parameter instability. A fund's betas shift with regime, so I would run rolling windows or a Kalman filter rather than assuming one constant loading over ten years. A single full-sample regression on a manager who changed style is a meaningless number.
Where candidates lose it
Reaching straight for robust standard errors as a universal fix. Robust errors do nothing about omitted variables or endogeneity, which is where the real inference error lives in fund regressions. Separate 'the coefficient is wrong' from 'the standard error is wrong' out loud, and mention stale pricing if the fund holds anything illiquid.
Expect next
- How would you detect stale pricing in a fund's returns?
- What would you do with only 36 monthly observations?
- How do you test whether the betas are stable?
Reported by candidates at AQR Capital Management (Investments, Greenwich, 2022). 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.