Quant interview questions: the 113 topics they test
Technical questions in a quant interview come back to these 113 topics, from 16 guides of the Academy. Each one is the rule to know in one sentence, linked to its guide: the why, a worked example and a timed quiz are there.
Probability and brainteasers
The probability tools quant and trading interviews test: conditioning and Bayes, counting, linearity of expectation, dice games, waiting times and uniform draws.
- Conditioning and Bayes. A conditional probability counts only the outcomes consistent with what you are told, and Bayes' rule weights each hypothesis by its prior.
- Counting. Combinations C(n, k) count unordered choices, and 'at least one' is easiest as 1 minus the chance of none.
- Linearity of expectation. The expectation of a sum is the sum of the expectations, even when the parts are dependent, so counting becomes a sum of indicator probabilities.
- Dice: maxima, variance and games. The maximum of several dice comes from P(max ≤ k), and the variances of independent dice add.
- Waiting for coin patterns. The expected wait for a pattern of flips depends on how a miss sets you back: heads-tails takes 4 flips, heads-heads 6.
- Uniform draws and geometric probability. For independent uniform draws on [0, 1], probabilities are areas in the unit square, and the maximum of 2 draws averages 2/3.
- Binomial counts and Kelly bets. A binomial count with n trials and probability p has mean n × p and variance n × p × (1 − p).
- Bayes in odds: each signal multiplies. In odds form, Bayes' rule multiplies the prior odds by the likelihood ratio, so independent pieces of evidence simply multiply their ratios.
Statistics and machine learning
The statistics quant interviews lean on: regression and its traps, correlation, what significance does and does not mean, normal tails, return arithmetic and the basics of model fitting.
- Regression and its traps. The OLS slope of y on x is the correlation times σy divided by σx, and with one regressor R² is the squared correlation.
- Correlation, variance of sums and spurious links. The variance of a sum adds the variances and twice the covariance, and correlations of trending price levels are often spurious, so correlate returns.
- Significance, errors and intervals. A p-value is the probability of data at least this extreme if the null hypothesis is true, and a mean's standard error is σ ÷ √n.
- Normal quantiles, skewness and fat tails. For a normal variable about 68.3% lies within 1 standard deviation and 95.4% within 2, and 1.645 is the one-sided 5% quantile.
- Return arithmetic and robust statistics. A Sharpe ratio scales with the square root of time, because mean returns grow with time and volatility with its root.
- Fitting models: overfitting, regularisation, metrics. A model that fits training data far better than new data has overfitted, and Lasso can set coefficients to zero while ridge only shrinks them.
- The Sharpe ratio luck alone would find. A Sharpe ratio estimated over T years has a standard error of about the square root of 1 ÷ T for a true Sharpe near zero.
Brainteasers
The classic puzzles and the one idea behind each: count the information a weighing gives, set up the right equation, work backwards, and never trust the fast answer.
- Weighings and information. Each weighing on a two-pan balance has 3 outcomes, so k weighings can tell apart at most 3k cases.
- Clock hands. The hour hand moves 0.5° a minute and the minute hand 6°, so at H:M the angle is |30 × H − 5.5 × M|.
- Handshakes, squares and paths. n people shaking hands once make n × (n − 1) ÷ 2 handshakes, and the shortest paths across an a × b grid number C(a + b, a).
- Number tricks: zeros, last digits, sums, divisors. The trailing zeros of n! equal its factors of 5, counted once per multiple of 5, again per multiple of 25, and so on.
- Equations that beat the fast answer. Puzzles built to tempt a fast answer are solved by writing the equation: an average speed over equal distances is the harmonic mean.
- Strategy puzzles: bridges, eggs, horses, doors. Strategy puzzles reward using every piece of information, such as switching doors when the host's choice carries information.
- Egg drops: count the floors a number of drops covers. With e eggs and d drops you can test up to C(d, 1) + C(d, 2) + … + C(d, e) floors.
Coding and algorithms
Predict what tricky Python prints, spot copies and shared references at a glance, and give the complexity of any loop or data structure before the interviewer finishes the question.
- Shared references, copies and mutable defaults. Assignment in Python copies the reference, not the object, so b = a shares the same list, while a[:] or .copy() makes a new one.
- Closures, scope, generators and in-place methods. A closure reads its variables when called, so lambdas made in a loop all see the last value unless bound with lambda i=i: i.
- Numbers, slices and truthiness. In Python 3, / always returns a float and // floors, keeping a float if either side is one.
- Counting loops and recursion. Count how far each loop's counter moves: a counter that doubles runs about log₂ n times, and independent nested loops multiply.
- Choosing the right data structure. Membership tests and lookups cost O(1) on average in a set or dict but O(n) in a list.
- Searching, sorting and the top k. A binary search must move a bound past the middle, lo = mid + 1, or it can loop for ever.
- Why append is O(1) even though it copies. A dynamic array that grows its capacity by a fixed proportion when full makes each append cost O(1) amortised.
Probability, advanced
The harder probability questions trading and quant firms ask: overlapping patterns, ruin and random walks, Poisson clocks, conditional means, records, derangements and stopping rules.
- Patterns and waiting times. The expected wait for a coin-flip pattern sums 2k over every k for which its first k flips equal its last k.
- Random walks, ruin and Markov chains. In gambler's ruin with win probability p, the chance of reaching N before 0 from k is (1 − rk) ÷ (1 − rN), with r = q ÷ p.
- Poisson processes and exponential clocks. A Poisson process with rate λ has no arrival in a period t with probability e−λt, and its waiting times are exponential and memoryless.
- Conditional means and the faces trick. Conditioning restricts the outcomes before averaging, and symmetry often gives the answer at once.
- Covariance and correlation. Cov(X, Y) = E[XY] − E[X] × E[Y], so a variable can be uncorrelated with its own square.
- Random draws: e, records, derangements and stopping. The number of uniform draws needed for their sum to exceed 1 averages e, and the share of derangements tends to 1 ÷ e.
- Symmetry and geometric puzzles. 3 random points on a circle form a triangle containing the centre with probability 1/4, as they lie on a semicircle with probability 3/4.
- How long a fair game lasts: distance squared. A fair one-step game started at k and stopped at 0 or N lasts k × (N − k) rounds on average.
Time series and econometrics
Work with AR and MA models, spot unit roots and spurious regressions, test cointegration for pairs trades, model volatility clustering and avoid the backtest traps.
- AR(1): mean, variance, forecasts and half-life. An AR(1) process x_t = c + φ x_{t−1} + ε_t with |φ| < 1 has a long-run mean of c ÷ (1 − φ).
- Reading autocorrelation: ACF, PACF and tests. An MA(q) process's autocorrelation function (ACF) cuts off after q lags, and an AR(p)'s partial autocorrelation function (PACF) after p lags.
- Unit roots and stationarity. When φ = 1 in an AR(1) the series has a unit root: a random walk whose shocks are permanent and whose variance grows with time.
- Cointegration and spurious regressions. 2 non-stationary series are cointegrated when a linear combination of them is stationary, which the Engle-Granger method tests on the regression's residual.
- Volatility clustering: GARCH and EWMA. Volatility clustering, large moves following large moves, is modelled on the variance with GARCH, whose long-run variance is ω ÷ (1 − α − β).
- Backtest pitfalls and forecast errors. Overlapping return windows make observations autocorrelated, so standard errors look too small unless corrected.
- How a GARCH forecast slopes back after a shock. A GARCH(1,1) variance forecast k days ahead is the long-run variance plus today's gap to it shrunk by (α + β) to the power k.
Machine learning for finance
Build models that survive real markets: catch overfitting and leakage, validate in time, use trees and boosting well, judge rare-event models with the right metrics and watch for drift.
- Overfitting and how to fight it. A model far better in sample than out of sample has learned noise, so keep a test set never used for any decision.
- Leakage and validation in time. Leakage lets information from the test period or the outcome reach the model, so every feature must be known before the prediction time.
- Random forests and boosting. A random forest averages deep trees grown on bootstrap samples with random feature subsets, while gradient boosting adds shallow trees fitting the errors so far.
- Rare events and the right metrics. On imbalanced data accuracy misleads, so a rare-event model is judged on precision, recall and AUC.
- Distances, explanations and drift. Distance-based methods such as k-nearest neighbours and k-means need standardised features and suffer in high dimensions, where points become roughly equally far apart.
- Why more trees stop helping: the correlation floor. The variance of an average of B trees with variance σ² and pairwise correlation ρ is ρσ² + (1 − ρ)σ² ÷ B.
Options and the Greeks
Use put-call parity, hedge delta, earn or pay gamma and theta, read vega and skew, and draw any payoff from calls, puts and spreads.
- Put-call parity. For European options with the same strike and expiry, a call minus a put equals the stock minus the present value of dividends and strike.
- Delta and delta hedging. A position's delta is the sum of each option's delta times its signed quantity, so short calls add negative delta and short puts positive.
- Gamma and theta. Gamma is how fast delta changes with the stock, so a delta-hedged long-gamma position gains from realised moves and pays theta.
- Vega, rho and what drives option value. Vega grows with time to expiry and peaks near the money, and an at-the-money option is worth about 0.4 × stock × volatility × √(time in years).
- Calendar spreads and risk reversals. A calendar spread long the longer option and short the shorter one is long vega, short gamma and long theta.
- Strikes and payoffs: spreads, straddles, collars. A spread's maximum loss is its net premium and its maximum profit the gap between strikes minus that premium.
- Early exercise and pin risk. Exercising an American call early throws away its time value, so it pays only just before a dividend larger than that time value.
- Gamma against theta: realised volatility against implied. A delta-hedged long option breaks even when the stock moves by its implied daily move, so it profits only if realised volatility beats implied.
Stochastic calculus
Handle Brownian motion, Itô's lemma, martingales and the change to the risk-neutral measure: the toolkit behind every derivatives desk, with the answers interviewers expect in seconds.
- Brownian motion: variances, covariances, scaling. A standard Brownian motion starts at 0 and has independent increments, with W_t − W_s normal of mean 0 and variance t − s.
- Itô's lemma and stochastic integrals. Because (dW)² = dt, Itô's lemma gives df(W) = f'(W) dW + ½ f''(W) dt, so d(W²) = 2W dW + dt.
- Martingales, stopping times and hitting problems. W_t, W_t² − t and eθW_t − θ²t/2 are martingales, so W from 0 hits b before −a with chance a ÷ (a + b).
- Risk-neutral pricing, Girsanov and Black-Scholes. Under the risk-neutral measure every discounted price is a martingale, so a stock without dividends drifts at r and μ never enters Black-Scholes.
- Mean reversion and jumps. The Ornstein-Uhlenbeck process dX = θ(μ − X) dt + σ dW pulls X towards μ at speed θ, with a long-run variance of σ² ÷ (2θ).
- Volatility drag: why more leverage can grow less. A log price grows at μ − σ²/2, so leverage L multiplies the expected return by L but the drag by L².
Market-making and trading games
Play the trading interview games well: centre a market on expected value, read what each trade tells you, manage your inventory, and size bets with Kelly.
- Making a market. A two-sided market is a tight spread centred on your expected value, and you buy on the other side's offer and sell on its bid.
- Adverse selection and the winner's curse. A market maker earns its spread on uninformed flow, loses to informed flow, and moves or widens its quotes when one side keeps trading.
- Managing inventory. A market maker long inventory lowers both its bid and its offer, and one that is short raises both.
- Expected values of dice and cards. Expected values add, and multiply for independent variables, but the expectation of a square is not the square of the expectation.
- Waiting times and optimal stopping. The number of failures before a first success with probability p averages (1 − p) ÷ p.
- Edge, fees and variance. A bet's edge is its expected value per unit staked, and fees are paid on every leg of a round trip.
- Kelly sizing. The Kelly criterion stakes the fraction (b × p − q) ÷ b of the bankroll, the edge divided by the odds.
- Where the break-even quotes sit when some traders know. A market maker sets its offer at the expected value given a buy and its bid at the expected value given a sell.
Estimation and market sizing
Answer any 'how many' or 'how big' question in a minute: break it into factors, divide demand by capacity, cross-check, and turn wild claims into numbers that expose them.
- The method: factors, demand and capacity. A Fermi estimate breaks the question into factors you can guess, multiplies them on paper, then checks the answer with a second, independent method.
- Sizing a market and testing a claim. Size a market bottom-up from real buyers and what they pay, and test the least reliable input first, usually the share who would actually buy.
- Anchors worth knowing: people, births and days. Good estimates start from a few anchors, such as populations, births and the right number of days, then apply shares, prices and frequencies.
- Little's law, flows and packing. Little's law says the number in a system equals the arrival rate times the time each spends in it.
- Money shortcuts. Money doubles in about 72 ÷ the yearly return in years, and a flat cash flow forever is worth that flow ÷ the required return.
- Market numbers: turnover, stake building and impact. Building a stake under a participation limit takes the shares needed divided by that share of daily volume, in days.
- Why Fermi estimates work: errors cancel. With n independent factors each uncertain by a factor k, the product is typically off by about k to the power √n.
Volatility trading
Know what a delta-hedged option really bets on, convert volatility between days and years, read skew and term structure, and trade variance swaps, dispersion and the VIX.
- Implied against realised: what a hedged option earns. A delta-hedged option position earns or loses the gap between the volatility the stock realises and the implied volatility paid or received.
- Volatility arithmetic. Volatility scales with the square root of time, so annual volatility is daily volatility times √252.
- Skew, smile and term structure. Equity index skew makes out-of-the-money puts dearer in volatility than calls, and implied volatility's term structure inverts in a crash.
- Variance and volatility swaps. A variance swap pays its variance notional times realised variance minus the strike squared.
- Correlation and dispersion. Index volatility depends on its members' volatility and their correlation, so selling index volatility against buying single-stock volatility bets that correlation falls.
- The VIX and its futures. The VIX measures the S&P 500's 30-day implied volatility from a strip of options, and its futures are usually in contango.
- Skew changes the delta you should hedge. When implied volatility rises as an index falls, an option's true delta is its model delta plus vega times the volatility change per spot point.
Option pricing
Price options 3 ways, with trees, Black-Scholes and Monte Carlo, and check any answer in your head with no-arbitrage bounds, quick rules and the Greeks.
- One-step and two-step trees. A binomial tree prices an option at the discounted average of its payoffs, weighted by q = (S(1 + r) − S_down) ÷ (S_up − S_down).
- Cox-Ross-Rubinstein trees and American options. An American option is priced backwards, taking at each node the higher of immediate exercise and the discounted value of continuing.
- Black-Scholes, forwards and quick bounds. Black-Scholes handles a dividend yield by replacing S with S e−qT, and an at-the-money call is worth about 0.4 × S × σ × √T.
- Greeks by finite differences and exotic options. Greeks come from bumping the price, with delta = (V_up − V_down) ÷ 2h and gamma = (V_up − 2V + V_down) ÷ h².
- Monte Carlo and finite-difference grids. A Monte Carlo price averages discounted payoffs over simulated paths, and its standard error falls with the square root of the number of paths.
- Option prices across strikes are probabilities. A tight call spread divided by its width prices the chance of ending above the strike, so call prices contain the risk-neutral distribution.
Interest rate modelling
Go from discount factors to forwards, par swap rates and PV01, switch between compounding conventions, price caps and swaptions, and know why desks use Hull-White and normal volatility.
- Discount factors, forwards and bootstrapping. A discount factor is today's value of 1 paid at a date, (1 + z)−T for an annual zero rate z.
- Swaps, PV01 and FRAs. A par swap rate makes the fixed leg worth the floating one: (1 − DF_n) ÷ (DF₁ + … + DF_n) with annual payments.
- Compounding conventions and SOFR. The same growth is (1 + r)T annually or er_c T continuously, with r_c = ln(1 + r), always below r.
- OIS discounting, futures and basis. Since 2008, collateralised derivatives are discounted at the overnight index swap rate that cash collateral earns.
- Short-rate models: Vasicek, Ho-Lee, Hull-White, CIR. Hull-White is Vasicek with a time-dependent drift chosen to fit today's yield curve exactly, and Ho-Lee fits it too without mean reversion.
- Caps, floors, swaptions and normal volatility. A cap is a strip of caplets, each a call on one period's rate paid at the period's end.
- Pricing a swaption in your head. In the normal model an at-the-money swaption is worth about 0.4 × normal volatility × √(expiry) × annuity × notional.
Python for data
Read NumPy and pandas code the way a desk quant does: views and copies, axes and broadcasting, NaN everywhere, joins that add rows, and backtests that do not peek at tomorrow.
- NumPy: views, axes, broadcasting and masks. A NumPy slice is a view, so changing it changes the original unless you call .copy(), while a list slice is a copy.
- NaN: missing values that spread. NaN is a float that equals nothing, itself included, so test it with np.isnan or pd.isna.
- pandas indexing and assignment. Select and assign in a single .loc call, df.loc[df.a > 0, 'b'] = 1, because chained indexing writes into a temporary copy.
- Merges, counts and speed. A left merge repeats each left row once per matching right row, so duplicate right keys add rows.
- Time series in pandas: returns, windows and dates. Shift a signal with .shift(1) before multiplying by returns, and compound returns with (1 + r).cumprod().
- Python traps that come up with data. Python's // rounds towards minus infinity and % takes the divisor's sign, while == compares values and is compares identity.
- Why NumPy is fast: memory layout. A NumPy float64 array stores each number in 8 contiguous bytes, while a Python list stores an 8-byte pointer to a separate 24-byte float object.
Software engineering in practice
Write code that a bank can run in production: safe threads, reliable tests, the Java and Python traps that cause real incidents, and the habits reviewers look for.
- Threads, races, deadlocks and the GIL. An unguarded counter++ shared by threads loses updates because it reads, adds and writes in 3 steps, so use a lock or an AtomicInteger.
- Tests that are fast, deterministic and useful. Unit tests should be fast and deterministic, so external services are mocked and tested apart, and the clock and time zone are injected.
- Java, Python and C++ traps behind real incidents. Equal Java objects must return equal hash codes or HashMap lookups miss, and an ever-growing static map leaks memory despite garbage collection.
- The cost of everyday operations. list.pop(0) shifts every other item, O(n), while deque.popleft() is O(1).
- Shipping safely: secrets, git, versions and flags. A live API key pushed to a shared branch must be revoked and rotated before any clean-up.
- Design, text encoding, sensitive data and HTTP. A factory behind a shared interface adds formats without touching callers, and PUT is idempotent, while POST creates again.
- Amdahl's law: the serial part caps the speed-up. A job whose share p runs in parallel on N cores takes at best (1 − p) + p ÷ N of its original time.
Say them out loud
The technical round of the video interview practice asks 5 questions on your desk, 90 seconds each, and reviews every answer. Each guide ends with a timed quiz.
Every topic comes from the Academy guides of this desk. Updated .