Quantitative Analyst Interview Questions (Algorithms & Python)

Black-Scholes is a mathematical model for pricing European-style options. It provides a theoretical price based on the underlying asset’s current price, the option’s strike price, time to expiration, risk-free interest rate, and volatility. The model derives a closed-form solution by constructing a risk-free hedge between the option and underlying asset.

Key assumptions include: the underlying follows geometric Brownian motion with constant volatility, no dividends, no transaction costs or taxes, continuous trading is possible, the risk-free rate is constant, and no arbitrage opportunities exist. Real markets violate these assumptions. Volatility varies over time and across strike prices (the “volatility smile”). Markets have transaction costs and liquidity constraints. Despite limitations, Black-Scholes provides a foundational framework and common language for options trading. Practitioners use it as a starting point, adjusting for real-world factors.

Monte Carlo simulation uses repeated random sampling to estimate outcomes that are difficult to solve analytically. For complex financial instruments, you simulate thousands of possible price paths for underlying assets, calculate the instrument’s payoff under each path, and average the results (discounted appropriately) to estimate fair value.

Monte Carlo is particularly valuable for path-dependent options like Asian options (averaging) or barrier options, for instruments with multiple underlying assets, and for risk calculations like VaR where you need the full distribution of potential outcomes. The method is flexible and can incorporate complex features like stochastic volatility or jumps. Drawbacks include computational intensity and the need for careful variance reduction techniques to achieve accurate estimates efficiently. Understanding when Monte Carlo is necessary versus when analytical solutions or simpler numerical methods suffice is an important quant skill.

Risk-neutral pricing is a framework for valuing derivatives where we price as if investors were indifferent to risk. Under this approach, all assets earn the risk-free rate in expectation, and we discount expected payoffs at the risk-free rate. This doesn’t mean investors actually are risk-neutral; rather, it’s a mathematical convenience that produces correct market prices.

The framework works because we can construct hedging portfolios that eliminate risk. Since hedged positions should earn the risk-free rate to prevent arbitrage, we can value derivatives by calculating expected payoffs under a risk-neutral probability measure and discounting at the risk-free rate. The actual (real-world) probability of outcomes doesn’t directly enter the pricing formula. This elegant result underlies most derivative pricing theory, though practitioners must remember that model risk and hedging imperfections in real markets mean theoretical prices may differ from realizable values.

Backtesting evaluates a trading strategy or model by testing it on historical data. You apply your rules to past data to see how they would have performed. This helps assess whether a strategy has predictive value and identifies potential weaknesses before risking real capital.

However, backtesting has significant pitfalls. Overfitting occurs when a model is tuned too closely to historical data, capturing noise rather than genuine patterns. Such models fail on new data. Look-ahead bias occurs when backtests inadvertently use information that wouldn’t have been available at the time. Survivorship bias occurs when testing only on assets that still exist, ignoring those that failed. Transaction costs, market impact, and liquidity constraints may make backtest profits unrealizable. Robust backtesting requires out-of-sample validation, realistic assumptions about execution, and skepticism about results that seem too good.

Let E be the expected number of flips to get two consecutive heads. From the start, flip once. With probability 0.5, you get tails and restart (having used 1 flip). With probability 0.5, you get heads. Given you got heads, flip again. With probability 0.5, you get heads again and you’re done (2 flips total). With probability 0.5, you get tails and restart (having used 2 flips).

Setting up the equation: E = 0.5(1 + E) + 0.5[0.5(2) + 0.5(2 + E)]. This simplifies to E = 0.5 + 0.5E + 0.5 + 0.25E, so 0.25E = 1.5, giving E = 6. The expected number of flips is 6. This type of recursive reasoning is common in quant interviews because it tests both probability intuition and the ability to set up and solve equations.

At 3:15, the minute hand points exactly at the 3, which is 90 degrees from 12. The hour hand, however, is not exactly at 3. It has moved 15 minutes worth of the way from 3 toward 4. The hour hand moves 360 degrees in 12 hours, or 0.5 degrees per minute. In 15 minutes, it moves 7.5 degrees past the 3 position.

The 3 position itself is 90 degrees from 12 (since each hour represents 30 degrees). So the hour hand is at 90 + 7.5 = 97.5 degrees from 12. The minute hand is at 90 degrees from 12. The angle between them is 97.5 – 90 = 7.5 degrees. These clock problems test attention to detail and the ability to break problems into components, skills essential for quantitative work.

This Fermi estimation tests structured thinking under uncertainty. Start with the US population: a few hundred million. Estimate vehicles: on the order of a few hundred million registered vehicles. Estimate average fuel consumption: if each vehicle fills up once per week, that’s 280 million fill-ups weekly.

Estimate gas station capacity: assume each station has 8 pumps and each pump can serve perhaps 4 cars per hour during busy periods. Operating maybe 16 peak hours daily, that’s about 500 fill-ups per day per station, or 3,500 per week. Dividing estimated weekly fill-ups by estimated station capacity gives a figure on the order of tens of thousands of gas stations. Public estimates are in the same broad order of magnitude, which is good enough for a Fermi approach. The key is showing logical structure, reasonable assumptions, and the ability to quickly calculate. Interviewers care more about your process than exact accuracy.

Value at Risk estimates the maximum potential loss a portfolio could face over a specified time period at a given confidence level. For example, a one-day VaR at a chosen confidence level is often read as: “there is a high probability the portfolio won’t lose more than a stated amount over one day.” VaR provides a single number summarizing downside risk that’s useful for risk limits and regulatory reporting.

Limitations are significant. VaR doesn’t describe the magnitude of losses beyond the threshold. It assumes historical patterns predict future risk, which fails during regime changes. Most VaR implementations assume normal distributions, underestimating fat-tail events. VaR can be gamed by taking risks that appear small under normal conditions but have catastrophic tail outcomes. I supplement VaR with Expected Shortfall (average loss beyond VaR), stress testing, and scenario analysis to capture risks VaR misses.

I start with a hypothesis based on economic intuition or observed market behavior. What inefficiency exists and why might it persist? Then I gather relevant data and perform exploratory analysis to see if the pattern exists. If promising, I formalize the strategy with specific rules for entry, exit, and position sizing.

Rigorous backtesting comes next, with careful attention to avoiding look-ahead bias, accounting for transaction costs, and testing across different time periods and market conditions. I validate on out-of-sample data and stress test under adverse scenarios. Before live trading, I run paper trading to verify execution assumptions. Once live, I monitor performance against expectations and have predefined criteria for when to stop or modify the strategy. Throughout, I maintain skepticism because most apparent patterns are noise or historical artifacts rather than persistent alpha sources.