Equity Research interview preparation

Questions

100

Traced to a firm

72

Firms

45

Updated

September 2026

Asked at

All firmsMorningstar12Man Group6Balyasny Asset Management5BLBlackRock5FTFranklin Templeton5MSCI5Jefferies4CSCredit Suisse3Fidelity Investments3Moody's3Perella Weinberg Partners3Point723S&P Global3The Vanguard Group3WMWellington Management3Advent International2Apollo Global Management2Bank of America2Carlyle Group2DED.E. Shaw2Houlihan Lokey2HSBC2Piper Sandler2Sequoia Capital2SSState Street2Viking Global Investors2WBWilliam Blair2ACAQR Capital Management1BGBaupost Group1BMBNY Mellon1Centerview Partners1Coatue Management1Goldman Sachs1GSGuggenheim Securities1HWHarris Williams1Insight Partners1Invesco1Mizuho1Moelis & Company1MSMorgan Stanley1PIMCO1SCSchroders1Scotiabank1T. Rowe Price1TSTruist Securities1

Topic

All topicsResearch process9Stock pitch6Company analysis8Investment philosophy5Valuation14Modelling2Portfolio and risk8Macro8Sector knowledge2Accounting8Career and fit12Industry knowledge6Quantitative research1Sector: technology3Sector: consumer1Sector: healthcare1Sector: energy1Sector: financials2Sector: industrials1Case and estimation2

Level

AnyCoreIntermediateHard

Type

AnyTechnicalCaseFitBrainteaserMarket view

1. 010How do you build a model that is detailed enough to be useful but simple enough that you can cover a lot of companies?ModellingHardtechnicalBalyasny Asset ManagementEquity Research · New York · 2026

   Say this

   Model deeply only where the variance is. For most companies two or three line items drive the outcome, so those get detailed driver builds and everything else gets a margin assumption or a percentage of sales.

   Then walk it

   1. Identify the swing factors first. For a retailer it is same-store sales and gross margin. For a bank it is net interest margin and provisions. For a software company it is net retention and sales efficiency. Model those properly.
   2. Everything else goes to ratios: other opex as a percent of revenue, working capital as days, tax at the guided rate. Precision there adds nothing and costs you maintenance time.
   3. Standardise the template across the coverage universe so the same row does the same thing in every file. That is what actually makes 15 names maintainable, because updating a quarter becomes mechanical.
   4. Build it around the disclosure you will actually receive. If the company only reports two segments, a five-segment model will be broken every quarter.
   5. And keep a one-page output: the drivers, the earnings bridge versus consensus, and the valuation. If the summary tab tells the story, the depth underneath can stay limited.
   6. The test I would apply: can I update this model in 20 minutes on results day? If not, it is too complex to cover 15 names with.

   Where candidates lose it

   Saying you would build the most detailed model possible. On the buy side, model complexity is a liability. The insight being tested is that modelling effort should be allocated to variance, not spread evenly.

   Expect next

   • How many names can one analyst realistically cover?
   • What goes on your summary tab?
   • How do you update on results day?

   Reported by candidates at Balyasny Asset Management (Equity Research, New York, 2026). Source: Wall Street Oasis.

2. 011How would you hedge a name that does not have a close public comparable?Portfolio and riskHardsuperdayBalyasny Asset ManagementEquity Research · New York · 2026

   Say this

   Hedge the exposures rather than the company. Decompose the position into its factor risks, market beta, sector, style, currency, commodity input, then hedge each with whatever liquid instrument matches it.

   Then walk it

   1. Start by decomposing: run the stock against factor returns and see what it is actually exposed to. Often a 'unique' business is really a bundle of common exposures.
   2. Hedge the market beta with an index future, sized on the regression beta rather than one.
   3. Hedge sector exposure with the closest sector ETF, accepting that the fit is imperfect. An imperfect hedge that removes 60 percent of the variance is better than no hedge.
   4. Hedge the specific input if there is one: a fuel-exposed business can be partly hedged with the commodity, a foreign earner with FX forwards.
   5. Then accept and size for the residual. The leftover idiosyncratic risk is the part you are actually being paid for, so the honest answer is that you hedge what you do not have a view on and hold what you do.
   6. And the practical constraint on a multi-manager platform: the risk system will impose factor limits anyway, so the hedge is often not optional. Saying that shows you understand how these seats actually operate.

   Where candidates lose it

   Reaching for a single 'closest competitor' short. If there were a close comp the question would not have been asked. The expected answer is factor decomposition, and naming the residual idiosyncratic risk as the intended exposure.

   Expect next

   • What residual risk are you left with?
   • How would you size the position?
   • What factor limits would you expect to operate under?

   Reported by candidates at Balyasny Asset Management (Equity Research, New York, 2026). Source: Wall Street Oasis.

3. 017How would you value a bank?ValuationHardtechnicalPerella Weinberg PartnersFinancial Institutions Group · New York · 2026Man GroupEquity Hedge · Boston · 2019

   Say this

   Price to tangible book against return on tangible equity, plus a dividend discount or residual income model. You do not use enterprise value or EBITDA, because for a bank debt is raw material and interest is revenue.

   Then walk it

   1. The core relationship: a bank should trade around book value if its return on equity equals its cost of equity, above book if it earns more, below if it earns less. The regression of price to book against ROTE across a peer group is the single most useful chart in the sector.
   2. Use tangible book, stripping goodwill and intangibles, because that is the capital actually supporting the balance sheet.
   3. For an intrinsic value, use a dividend discount model or residual income, since dividends are constrained by regulatory capital and that constraint is the real driver of distributable cash.
   4. The forecast drivers are net interest margin, loan growth, fee income, the cost-to-income ratio, and the provision charge. Provisions are where the cycle shows up and where forecasts go wrong.
   5. Capital is the binding constraint on everything. CET1 ratio against the regulatory requirement determines whether the bank can grow, buy back stock or must raise equity, so I would model capital explicitly rather than treating it as an output.
   6. And the thing that actually breaks bank valuations: credit losses are non-linear. A small deterioration in the macro can wipe out several years of earnings, which is why banks trade below book in a downturn regardless of reported profit.

   Where candidates lose it

   Applying EV/EBITDA or a standard unlevered DCF. It is meaningless for a bank and it is an instant fail in a financials interview. Lead with price to tangible book versus ROTE and the reason enterprise value does not apply.

   Expect next

   • Why can you not use enterprise value?
   • What happens to the valuation if rates fall 200 basis points?
   • How do you forecast provisions?

   Reported by candidates at Perella Weinberg Partners (Financial Institutions Group, New York, 2026); Man Group (Equity Hedge, Boston, 2019). Source: Wall Street Oasis.

4. 019How do you assess earnings quality?AccountingHardtechnicalMoody'sCorporate Finance · New York · 2018MorningstarEquity Research · Chicago · 2023

   Say this

   Compare earnings to cash. If net income is consistently above cash from operations, something is being recognised that has not been collected. Then check the accruals, the adjustments and the one-offs.

   Then walk it

   1. The headline test: cash conversion. Cash from operations divided by net income, tracked over several years. Persistent divergence is the single best red flag available from published accounts.
   2. Then working capital. Receivable days rising faster than revenue means revenue is being pushed to customers or collection is deteriorating. Inventory days rising means a write-down is coming.
   3. Then the adjustments. Compare GAAP to the company's adjusted figures and see what is being excluded. Restructuring charges taken every year for five years are not one-off, they are operating costs in disguise.
   4. Then capitalisation choices: capitalised development costs, capitalised interest, and the depreciation life. Extending useful lives flatters earnings with no economic change.
   5. Then the tax rate and the below-the-line items, since a sudden drop in the effective tax rate can manufacture an EPS beat.
   6. For a note, the useful summary is a bridge from reported earnings to what I think the sustainable earnings power is, with each adjustment listed. That bridge is often the most valuable page in a research report.

   Where candidates lose it

   Listing ratios without the organising idea. The organising idea is that accounting earnings involve judgement and cash does not, so every test is a version of comparing the two. Say that first.

   Expect next

   • What is the single best red flag?
   • How do you treat stock-based compensation?
   • Walk me through a company you thought had poor earnings quality.

   Reported by candidates at Moody's (Corporate Finance, New York, 2018); Morningstar (Equity Research, Chicago, 2023). Source: Wall Street Oasis.

5. 020Should stock-based compensation be treated as a real expense?AccountingHardtechnicalTechnology coverageLong-only asset management

   Say this

   Yes. It is a genuine cost to existing shareholders even though no cash leaves the company, because it transfers ownership. Adding it back to get to adjusted EBITDA or free cash flow overstates what shareholders actually keep.

   Then walk it

   1. The economic argument: if the company paid those employees in cash and then issued shares to raise the same amount, nobody would argue the salary was not an expense. The two are identical in substance.
   2. It shows up as dilution. Share count creeps up every year, so per-share metrics deteriorate even when totals look fine. That is the cost, and it is real.
   3. Companies obscure it by buying back stock to offset dilution and then describing the buyback as capital return. It is not; it is paying cash for compensation already granted.
   4. The practical treatment I would use: expense it fully in the earnings I value, and if I want a cash-based measure, subtract the buyback needed to keep share count flat rather than adding SBC back.
   5. The counterargument worth acknowledging: the accounting charge is based on grant-date fair value, which can be a poor estimate of the eventual cost, and the expense is lumpy. So the number is imperfect even if the principle is clear.
   6. In practice this matters most in software, where SBC can be 15 to 25 percent of revenue. Whether you expense it decides whether a company is profitable at all.

   Where candidates lose it

   Accepting the company's adjusted figure because it is what consensus uses. Research is supposed to be the check on that. Have a view, and know roughly how large SBC is as a percentage of revenue for the sector you claim to follow.

   Expect next

   • How large is it as a percentage of revenue in software?
   • How do you handle it in a DCF?
   • What does that do to the sector's valuation?
6. 029How do you construct a factor, and how would you decide whether it is real?Quantitative researchHardtechnicalACAQR Capital ManagementInvestment Research · New York · 2021

   Say this

   Define the signal, rank the universe on it, form long-short portfolios from the extremes, and measure the spread after controlling for known factors. It is real only if it survives transaction costs, out-of-sample testing and an economic explanation.

   Then walk it

   1. Construction: choose the metric, neutralise for size, sector and region so you are not just picking up a sector bet, then sort into quantiles and go long the top and short the bottom with rebalancing at a defined frequency.
   2. Measure the spread return, its volatility, the information ratio, and the turnover it requires. Turnover matters because a signal that needs daily rebalancing can be profitable on paper and unprofitable after costs.
   3. Control for known factors. If your new signal's returns disappear once you regress against value, momentum, quality and size, you have rediscovered an existing factor with a new name.
   4. Then the tests that actually matter: out-of-sample and out-of-region performance, stability across sub-periods, and how many specifications you tried before finding this one. Data mining is the default explanation for any new factor.
   5. And demand an economic story. A factor should be compensation for a risk, or exploitation of a behavioural bias, or a structural constraint on other investors. Without that, decay after publication is the base case.
   6. The honest position: most published factors do not survive replication, so the prior on any new one should be skeptical.

   Where candidates lose it

   Describing the mechanics with no discussion of multiple testing and data mining. The intellectual content of modern factor research is that backtests are easy and robustness is hard. Say so.

   Expect next

   • How would you optimise the construction?
   • Why do factors decay after publication?
   • How would you know if you had overfitted?

   Reported by candidates at AQR Capital Management (Investment Research, New York, 2021). Source: Wall Street Oasis.

7. 031How do you decide when to sell?Portfolio and riskHardsuperdayAsset managementHedge funds

   Say this

   Three reasons and only three: the thesis played out and the price reflects it, the thesis is broken, or something better came along. Never sell because the price fell, and never hold because you are down.

   Then walk it

   1. Thesis achieved: the variant view became consensus and the upside to your revised target is no longer compelling. This is the happy case and people systematically sell too early here.
   2. Thesis broken: the specific thing you said would happen did not, or a fact you relied on turned out false. This should trigger a sale regardless of price, and it is where writing down the falsifier in advance pays for itself.
   3. Better use of capital: opportunity cost. In a concentrated portfolio every new idea must displace something, which imposes useful discipline.
   4. What is not a reason: the price fell, so it is cheaper now. That is only a reason to buy more if the thesis is intact, and only if you have checked rather than assumed.
   5. The behavioural safeguards: a written thesis with falsifiers, a scheduled review after every result, and a rule that you re-underwrite a position from scratch rather than defending your existing note.
   6. And separate trimming from selling. Reducing on valuation while the thesis compounds is a different decision from exiting, and conflating them is how people sell their best ideas.

   Where candidates lose it

   Giving a price-based rule like a fixed stop loss as the whole answer. For fundamental investing, the sell decision is thesis-based. Stops are a risk management overlay, not a research judgement, and saying only that reveals a trader's frame in a research seat.

   Expect next

   • How do you avoid selling winners too early?
   • Do you use stop losses?
   • How do you re-underwrite a position?
8. 032How do you size a position?Portfolio and riskHardsuperdayHedge fundsAsset management

   Say this

   By conviction and by downside, not by expected upside. The question is how much you lose if you are wrong, multiplied by how likely that is, against the portfolio's tolerance for that loss.

   Then walk it

   1. Start from the downside. If the bear case is minus 40 percent and you would be uncomfortable losing more than 2 percent of the fund on one name, the position is capped at about 5 percent.
   2. Then conviction, which really means how confident you are in the analysis and how falsifiable it is. A thesis with a clear near-term test supports a larger position than one that depends on a five-year structural view.
   3. Then correlation. Three positions expressing the same macro view are one position. Sizing has to be done at the portfolio level or you accumulate hidden concentration.
   4. Then liquidity: how many days of average volume is the position, and can you exit it in a stressed market? Illiquidity is a real constraint on size regardless of conviction.
   5. The Kelly criterion is the theoretical frame, but full Kelly is far too aggressive in practice because you cannot estimate probabilities that precisely. Most investors run a fraction of it, and saying that shows you know the theory and its limits.
   6. In a multi-manager seat, most of this is imposed by the risk system anyway, and the analyst's job is to argue for the sizing within those limits.

   Where candidates lose it

   Sizing by upside. Everyone's best idea has the most upside, and sizing on that alone is how funds blow up. Downside and correlation are the content of a real answer.

   Expect next

   • What is your maximum position size?
   • How do you handle correlated positions?
   • Would you add to a loser?
9. 037How would you model a subscription software business, and what metrics matter?Sector: technologyHardtechnicalInsight PartnersSoftware · New York · 2022Piper SandlerInvestment Banking · Burlingame · 2026

   Say this

   Model the recurring revenue base by cohort rather than the income statement. Opening ARR, plus new, plus expansion, less churn and downgrades, gives closing ARR. Everything else follows from that roll-forward.

   Then walk it

   1. The ARR bridge is the model. Once you have opening ARR, new bookings, expansion and churn, revenue is largely determined, because recognised revenue is a lagging function of the contracted base.
   2. Key metrics: net revenue retention, gross retention, gross margin, customer acquisition cost payback, and the rule of forty which is growth plus free cash flow margin.
   3. Net retention above 110 percent is the single most important number, because it means the installed base grows without selling anything new. That is what justifies a high revenue multiple.
   4. Watch the gap between billings, revenue and deferred revenue. Billings lead revenue, so a slowdown shows up in billings a quarter or two before it hits the reported line. That is often where the variant view lives.
   5. Cost side: gross margin tells you how much real compute or support sits in cost of revenue, sales and marketing efficiency tells you whether growth is bought or earned, and R&D as a share of revenue tells you about future product.
   6. And take stock-based compensation seriously, because in software it is large enough to determine whether the company is profitable at all.

   Where candidates lose it

   Modelling revenue directly and ignoring the ARR bridge and deferred revenue. Also quoting the rule of forty without knowing whether it uses free cash flow margin or operating margin, since the two give very different answers.

   Expect next

   • What is the rule of forty?
   • Why do billings lead revenue?
   • What net retention would justify a 10 times revenue multiple?

   Reported by candidates at Insight Partners (Software, New York, 2022); Piper Sandler (Investment Banking, Burlingame, 2026). Source: Wall Street Oasis.

10. 042How do you normalise earnings for a cyclical company?ValuationHardtechnicalFTFranklin TempletonOil and Gas · San Mateo · 2024

    Say this

    Estimate what the business earns through an average cycle, not at either extreme. Take mid-cycle volumes and mid-cycle margins, adjusted for any structural change since the last cycle, and value that.

    Then walk it

    1. Method one: average the margin over a full cycle, usually seven to ten years, and apply it to current revenue. Simple and defensible.
    2. Method two: estimate mid-cycle volume and mid-cycle price separately, then rebuild the income statement. More work, but it lets you adjust each independently.
    3. Method three: normalise on the balance sheet instead, using return on invested capital through the cycle applied to today's capital base. Useful when volumes have changed structurally.
    4. The critical adjustment: has anything structural changed since the last cycle? Capacity closures, consolidation, a new cost position, or demand substitution mean history is not a clean guide. This is where the analysis is.
    5. Then apply a mid-cycle multiple to the normalised figure. The common error is applying a peak multiple to normalised earnings, or a normalised multiple to peak earnings; the two must be consistent.
    6. And show the earnings range rather than a point. For cyclicals the honest output is a value at trough, mid and peak, with a probability view on where in the cycle we are.

    Where candidates lose it

    Normalising the earnings but not the multiple, or ignoring structural change and treating the last cycle's average as destiny. Consistency between the earnings base and the multiple is the whole discipline.

    Expect next

    • How do you know where in the cycle you are?
    • What has structurally changed in that industry?
    • Why do cyclicals look cheapest at the top?

    Reported by candidates at Franklin Templeton (Oil and Gas, San Mateo, 2024). 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.