Fermi Estimation for Quant Interviews: How Many Piano Tuners, Gas Stations, Trades

Q: What books should I use for Fermi practice?

Guesstimation by Lawrence Weinstein and John Adam is the standard book of Fermi problems with solutions. How Many Licks? by Aaron Santos is similar. Both have hundreds of problems; working through 30–50 builds the pattern recognition. Heard on the Street by Timothy Crack also has many Fermi-style trader problems with worked solutions. Practice problems verbally, out loud, with timing pressure to simulate interview conditions.

Fermi Estimation for Quant Interviews: How Many Piano Tuners, Gas Stations, and Trades Per Day?

Fermi estimation — the art of approximating quantities you don’t directly know by breaking them into parts you can roughly estimate — is a foundational skill at quant trading interviews. Jane Street, Optiver, SIG, Akuna, IMC, Citadel Securities, and most market-making firms ask Fermi-style questions either explicitly (“how many gas stations are in the US?”) or implicitly within market-making rounds (“make a market on the number of pages in War and Peace”). The skill probes whether you can anchor numerical estimates rapidly, decompose problems sensibly, and communicate your reasoning under time pressure.

Named after physicist Enrico Fermi, the technique is the foundation of how skilled traders form initial views on unfamiliar quantities. Strong Fermi reasoning is not about being right; it’s about being defensibly close while showing your work.

The Core Technique

Fermi estimation works by decomposing an unknown quantity into a product (or sum) of factors, each of which can be estimated within an order of magnitude. The structure:

Identify the quantity to estimate.
Decompose it into multiplicative components.
Estimate each component using base rates, common knowledge, or reasonable guesses.
Multiply through.
Sanity-check the result against any other intuitions you have.

Errors in individual components partially cancel when multiplied, so even rough per-component estimates often produce surprisingly accurate final answers.

Canonical Example: Piano Tuners in Chicago

Fermi’s original problem. How many piano tuners are in Chicago?

Decomposition:

Population of Chicago: ~3 million
Pianos per household: maybe 1 in 20 households has a piano. Households per person: ~3. So pianos per person ≈ 1 / (20 × 3) ≈ 0.017. Total pianos ≈ 3M × 0.017 ≈ 50,000.
Tuning frequency: maybe once per year per piano. So 50,000 tunings per year.
Tunings per tuner: working full-time, ~3 tunings per day, ~250 working days = ~750 tunings per year.
Number of tuners: 50,000 / 750 ≈ 67.

Actual answer: roughly 80–100 piano tuners in Chicago. The estimate is within 50% — an excellent Fermi result.

Common Quant Interview Versions

Gas stations in the US

Population: ~330M. Cars per person: ~0.8 (rough US ratio). So ~265M cars. Cars per gas station: gas stations need maybe 500 cars in their service area to be viable. So ~265M / 500 ≈ 530,000 gas stations. Actual: ~150,000. The estimate is high by ~3x — not great, suggesting the “cars per station” component was too low.

Better decomposition: people drive ~12,000 miles per year, ~25 mpg, so ~480 gallons per person per year. Total US: ~150 billion gallons. Average gas station: maybe 800,000 gallons per year. Stations: ~190,000. Closer to actual.

The interviewer doesn’t care that your first attempt was off; they care that you noticed and corrected.

Daily trading volume of Apple stock

Apple market cap: ~$3T. Average daily volume across major stocks: ~0.5% of market cap. So daily volume in dollars: ~$15B. At ~$200/share: ~75M shares. Actual range: 30–100M shares depending on the day. Within striking distance.

Number of options contracts traded daily on US exchanges

Total US equity market: ~$50T. Options market is roughly comparable in notional. Average contract represents 100 shares; average strike around $100 (depends heavily on the asset mix). So average contract notional ~$10,000. Daily options volume: depends on turnover… ~50M contracts per day in the US is the actual number; estimating from notional turnover gives a rough answer in the right order of magnitude.

How many M&Ms in a jar

Classic. Volume of jar (visual estimate). Volume of one M&M (~1 cm³). Packing density (~70% for spheres). Number ≈ jar volume × 0.7. The interviewer cares whether you set up the calculation, not the exact answer.

Common Patterns in Fermi Problems

Per-capita / per-household decomposition

For market-size questions: total US population × per-person frequency × demographic adjustment. This decomposition works for almost any consumer product or service.

Production-side decomposition

For supply-side questions: number of producers × output per producer. Useful when supply-side is more tangible than demand-side.

Time decomposition

For volume questions: events per unit time × time period. “How many flights per day from JFK?” = number of gates × flights per gate per day.

Bracket-and-narrow

If you’re unsure about a component, give a range (10–100) and propagate. Final answer becomes a range rather than a point estimate; this is honest about the uncertainty.

Anchoring Strategies

Fermi problems require knowing rough magnitudes for common quantities. Memorize a few base rates:

US population: ~330M; world population: ~8B
US households: ~125M; cars: ~270M
Square miles of US: ~3.8M; US GDP: ~$28T
Global stock market cap: ~$110T; US: ~$55T; daily US equity volume: ~$500B
Major US cities: NYC ~9M, LA ~4M, Chicago ~3M, Houston ~2M
Lifespan: ~80 years (US); working years: ~40
Workdays per year: ~250

These are anchors; you don’t need to memorize many more than these. From them you can derive most relevant quantities.

What Interviewers Look For

Decomposition over guessing

“I’d guess 100,000 gas stations” is much weaker than “330M people, ~500 cars per station, 530,000 stations” — even if both reach the same answer. Show your work; the structure is the answer.

Verbalize each step

Don’t compute silently. Say: “330M people. Maybe 0.8 cars per person. So 270M cars. Cars per station…” Strong candidates make their reasoning audible.

Sanity-check

“That gives 530,000, which feels high. Let me check from the supply side…” Recognizing when an estimate seems off and trying again is a strong signal.

Comfort with imprecision

Fermi answers are within an order of magnitude or two. Strong candidates don’t pretend to know more than they do; they confidently report rough estimates.

Speed

For interview Fermi questions, you have 1–3 minutes. Don’t paralyze on perfect components; pick reasonable values quickly and propagate.

Connecting Fermi to Market-Making

Market-making rounds are essentially Fermi questions where you also commit to a market. “Make a market on the number of pages in War and Peace” requires Fermi anchoring (probably 1,000–1,500 pages) plus commitment to a bid-ask spread reflecting your uncertainty.

Strong Fermi skills directly help market-making: you anchor faster and more defensibly, allowing you to focus interview energy on the trading dynamics rather than the estimation. See Market-Making Interview Questions for how Fermi feeds into market-making.

Frequently Asked Questions

How important is Fermi specifically vs general estimation?

Critical for market-making firm interviews. Jane Street, Optiver, SIG, Akuna, IMC explicitly ask Fermi questions; Citadel Securities and Two Sigma sometimes do; investment bank Strats interviews rarely do. If you’re targeting market-making firms, Fermi practice is essential. If you’re targeting derivatives-pricing or systematic equity research, less central but worth a few hours of practice.

What if I get the answer dramatically wrong?

If your decomposition is sound and your reasoning is clear, being off by 5x is acceptable. Being off by 100x suggests a missing or badly-estimated component; the interviewer will probe. Catch yourself before the interviewer does: as you’re about to commit to an answer, sanity-check against intuition. “That gives 5M; that seems too high; let me reconsider…” is a good move. Pretending to be confident when your answer is obviously off is a worse signal than catching the error.

Should I memorize lots of base rates?

Memorize a small canonical set (US population, GDP, daily equity volume, life expectancy, etc.) and derive others. Memorizing dozens of base rates (US healthcare spending, retail sales, etc.) doesn’t help much because the questions vary too much. Better to internalize the decomposition pattern and a few anchors.

How does Fermi reasoning differ from probability reasoning?

Fermi is about anchoring numerical estimates from incomplete information; probability is about quantifying uncertainty given a stochastic process. They overlap: market-making rounds combine both (anchor a fair value via Fermi; quote a spread reflecting probabilistic uncertainty). Strong candidates are fluent in both and don’t conflate them. Probability without Fermi gives clean math but no anchor; Fermi without probability gives an anchor but no spread.

What books should I use for Fermi practice?

Guesstimation by Lawrence Weinstein and John Adam is the standard book of Fermi problems with solutions. How Many Licks? by Aaron Santos is similar. Both have hundreds of problems; working through 30–50 builds the pattern recognition. Heard on the Street by Timothy Crack also has many Fermi-style trader problems with worked solutions. Practice problems verbally, out loud, with timing pressure to simulate interview conditions.