DRW Interview Questions 2026

100 Questions Unlock all questions ↓
Filter
Topic area
Difficulty

Filters apply to the free questions below.

Practice Bank

DRW Quant Interview Practice

↗

A small observatory keeps a text log of one sensor event every 10 seconds, running continuously for 10 days. Each event consists of a 1,000-byte text payload and a 100-byte header containing its timestamp and device identifier. The text compressor reduces every payload to 30% of its original byte count, but the headers are stored uncompressed. There are no other files, indexes, backups or filesystem overheads.

How many megabytes (MB) of disk space will the archive occupy? Use 1 MB = 1,000,000 bytes, not \(2^{20}\) bytes.

Hint 1

The compressor changes 1,000 bytes of text into 300 bytes but leaves the 100-byte header untouched.

Multiply the new 400-byte event size by the number of 10-second intervals in \(10\times24\times60\times60\) seconds, then use the specified decimal MB conversion.

There are \(24\cdot60\cdot60=86{,}400\) seconds in a day. Ten days contain 864,000 seconds, so a recorder writing one event every 10 seconds stores 86,400 events.

Each payload occupies \(0.30\cdot1{,}000=300\) bytes after compression. The header retains its full 100 bytes, giving \(300+100\)=400 bytes per event. Thus total storage is

\(86{,}400\times400\) = 34,560,000 bytes.

Under the required decimal convention this is 34.56 MB. Applying the compression percentage to the header as well would be incorrect: the 100-byte timestamp/identifier is explicitly excluded. If someone instead reported binary mebibytes, the same bytes would be about 32.96 MiB, but the question requests MB.

Assumptions

  • All retention days have 24 hours and logging occurs without gaps.

  • Each text payload shrinks to exactly 30% of its original size; metadata headers do not shrink.

  • No disk overhead is counted and MB is defined decimally.

Final Answer

34.56 MB (34,560,000 bytes).

Numerical reasoning and estimation
↗

At a science exhibit, two lights are controlled by a single fresh random number U drawn uniformly from 0 to 1. The blue light switches on when \(U<0.30\), and the amber light switches on when \(U<0.70\). A new U is drawn only for the next show: during one show the lights always use the same number.

What is the probability that both lights switch on? Also list the possible on/off combinations with their probabilities. Would you get the correct answer by multiplying the two individual activation probabilities?

Hint 1

If the blue light is on, what must be true of the amber light?

Partition the single uniform draw into \([0,0.30)\), \([0.30,0.70)\), and \([0.70,1]\). Each interval corresponds to one joint outcome.

Let \(B=\{U<0.30\}\) and \(A=\{U<0.70\}\). Because \(0.30<0.70\), B is a subset of A. Thus \(\Pr(B\cap A)=\Pr(B)=0.30\), not \(\Pr(B)\Pr(A)=0.21\).

On \(0\le U<0.30\), both lights are on, probability 0.30. On \(0.30\le U<0.70\), blue is off and amber is on, probability 0.40. On \(0.70\le U\le1\), both are off, probability 0.30. Blue on with amber off is impossible, probability zero. These four probabilities sum to one.

The product rule \(\Pr(B\cap A)=\Pr(B)\Pr(A)\) would require independence, which is contradicted by the deterministic inclusion \(B\subset A\). For instance \(\Pr(A\mid B)=1\) whereas \(\Pr(A)=0.70\).

Assumptions

  • U is a continuously uniform random draw on \([0,1]\).

  • The same U controls both thresholds for a given exhibit show.

Final Answer

\(\Pr(\text{both on})=0.30\). The (blue,amber) outcomes are (on,on): 0.30, (off,on): 0.40, (off,off): 0.30, and (on,off): 0. Multiplying \(0.30\times0.70\) is incorrect.

Probability foundations
↗

A conservator moves two specimen jars, chosen uniformly at random without replacement, from cupboard A into cupboard B. Before the move, A contains three jars marked fragile and three unmarked jars; B contains two fragile and four unmarked jars.

An assistant watching the move reports only that at least one of the two transferred jars was fragile. The conservator cannot see which jars were moved. Two jars are then selected uniformly without replacement from the now-enlarged cupboard B.

Given the assistant's report, what is the probability that both inspected jars are fragile?

Hint 1

Let \(K\) be the number of fragile jars transferred. The report rules out \(K=0\), but it does not tell you whether \(K=1\) or \(K=2\).

First compute \(P(K=1\mid K\ge1)\) and \(P(K=2\mid K\ge1)\) using combinations. Then, for each possible \(K\), count ways to draw two fragile jars from the eight jars now in B.

A transfer is an equally likely two-jar subset of A. There are \(\binom62=15\) such subsets, of which

\[ P(K=0)=\frac{\binom32}{15}=\frac15,\qquad P(K=1)=\frac{\binom31\binom31}{15}=\frac35,\qquad P(K=2)=\frac{\binom32}{15}=\frac15. \]

The report is \(R=\{K\ge1\}\), with probability \(4/5\). Bayes conditioning gives \(P(K=1\mid R)=3/4\) and \(P(K=2\mid R)=1/4\). When \(K=k\), the enlarged cupboard B has \(8\) jars, exactly \(2+k\) of them fragile. Drawing two without replacement has conditional probability

\[ P(\text{two fragile}\mid K=k)=\frac{\binom{2+k}{2}}{\binom82}. \]

Thus the required mixture is

\[ \frac34\frac{\binom32}{\binom82}+\frac14\frac{\binom42}{\binom82} =\frac34\frac3{28}+\frac14\frac6{28}=\boxed{\frac{15}{112}}. \]

An exact enumeration confirms the result: among the 12 transferable pairs satisfying the report, nine contain one fragile jar and three contain two. Each leads to 28 equally likely inspection pairs, with respectively three and six favourable ones. Therefore the answer is \((9\cdot3+3\cdot6)/(12\cdot28)=45/336=15/112\). Treating the two inspected jars as independent or replacing the hidden transfer by its average composition would give the wrong model.

Assumptions

  • Within each cupboard, all subsets of a given sampled size are equally likely.

  • Transfer and inspection are performed without replacement as stated.

  • The only transfer information disclosed is the event that at least one moved jar was fragile; the full composition remains hidden.

Final Answer

\(15/112\), approximately \(13.39\%\).

Conditional probability
↗

A charity sends a referral invitation to one volunteer. Anyone who receives the invitation independently forwards it to zero, one or two new people, with probabilities \(1/4,1/4,1/2\). The coordinator guarantees that recipients are distinct, so nobody can receive the invitation twice. The invitation chain stops after three waves of forwarding: the original volunteer creates wave 1, wave 1 creates wave 2, and wave 2 creates wave 3. People in wave 3 do not forward.

Preparing an invitation for each new person costs $4 in wave 1, $2 in wave 2 and $1 in wave 3. The original volunteer's invitation has already been paid for. What is the expected number of new recipients across the three waves, and the expected total preparation cost? Do not assume the chain is equally likely to reach every wave.

Hint 1

You need the expected wave sizes rather than the probability of any single complete family tree.

Calculate the mean number of people forwarded to by one recipient and use \(E[Z_{g+1}]=mE[Z_g]\) from \(Z_0=1\). Apply each wave’s own cost only after finding its expected size.

Let \(Z_g\) be the number of recipients in wave \(g\), counting the original volunteer as \(Z_0=1\). Each recipient independently produces \(X\) new recipients with mean
\[ m=E[X]=0\cdot\tfrac14+1\cdot\tfrac14+2\cdot\tfrac12=\tfrac54. \]
Conditioning on a current wave of size \(Z_g\), the next wave has expected size \(mZ_g\). Therefore the tower property gives \(E[Z_{g+1}]=mE[Z_g]\). Iterating yields
\[ E[Z_1]=\tfrac54,\qquad E[Z_2]=\tfrac{25}{16},\qquad E[Z_3]=\tfrac{125}{64}. \]
Linearity of expectation, which does not require independence between wave sizes, gives the expected number of new recipients:
\[ E[Z_1+Z_2+Z_3]=\frac{80+100+125}{64}=\boxed{\frac{305}{64}}. \]
Different waves have different prices, so the expected cost is
\[ 4E[Z_1]+2E[Z_2]+E[Z_3] =5+\frac{25}{8}+\frac{125}{64}=\boxed{\frac{645}{64}\text{ dollars}}. \]
The expected population grows because \(m>1\), but with only three waves both expected workload and cost are finite. Using a single averaged price per wave would be unjustified because the waves have different expected sizes.

Assumptions

  • All offspring counts are independent and have the same stated distribution.

  • Every contact is to a new, unique person; wave 3 has no descendants.

  • Only invitations to new recipients in waves 1, 2 and 3 incur the listed costs.

Final Answer

Expected new recipients: \(\frac{305}{64}\approx4.765625\). Expected preparation expenditure: \(\$\frac{645}{64}=\$10.078125\).

Expectation and expected value
↗

Two co-owners must agree before sending fragile camera equipment on a weather-sensitive expedition. The weather is good with probability 60% and bad otherwise. If they depart, each co-owner earns $200 in good weather but loses $200 in bad weather. If either votes to veto, the trip is cancelled and each earns $0. Voting is simultaneous.

Before voting, each may privately buy a weather report for $16.80, whether or not the trip ultimately departs. A report says good or bad and is correct with probability 80% in either weather state. The two reports are independent conditional on the actual weather. Neither the report purchase nor its contents are visible to the other co-owner before the votes.

Find a symmetric equilibrium in which each co-owner buys a report with some probability \(q\), vetoes after a bad report, and otherwise permits departure. Derive \(q\) and check that neither the report decision nor any of the voting decisions offers a profitable unilateral deviation. Why does the other person's veto policy affect the value of buying information?

Hint 1

A vote for departure only matters when the other owner also permits. Calculate the probability that the other owner permits in good weather and in bad weather as functions of q.

Compare the expected payoff from permitting without a report with the payoff from permitting only after a good report, minus the $16.80 report price. Equate these payoffs for an interior mixed purchase strategy, then check each signal's voting incentive.

A single veto cancels both owners' earnings, so the return to permitting depends on whether the other owner permits. Suppose the other owner buys with probability \(q\), permits without a report or after a good report, and vetoes after a bad report. Conditional on good weather, that owner permits with probability \((1-q)+0.8q=1-0.2q\). Conditional on bad weather, permission has probability \((1-q)+0.2q=1-0.8q\).

For an owner with no report, permitting therefore has expected trip payoff

\[U_0(q)=0.6(200)(1-0.2q)+0.4(-200)(1-0.8q)=40+40q.\]

Vetoing earns zero, so permitting is optimal for any \(q\in[0,1]\). If the owner buys a report, the unconditional joint contributions to expected payoff from permitting after each possible signal are

\[U_{+}(q)=0.6(0.8)(200)(1-0.2q)-0.4(0.2)(200)(1-0.8q)=80-6.4q,\]
\[U_{-}(q)=0.6(0.2)(200)(1-0.2q)-0.4(0.8)(200)(1-0.8q)=-40+46.4q.\]

Each is the probability-weighted payoff from the specified signal and the other person's permission; because the probability of each signal is positive, its sign also gives the correct conditional voting preference. Whenever \(q<40/46.4\), a positive report warrants permission and a negative report warrants veto. At such \(q\), the net value of buying and then voting optimally is \(U_+(q)-16.80=63.20-6.4q\). For genuinely mixed acquisition, both buying and not buying must be best responses:

\[63.20-6.4q=40+40q\quad\Longrightarrow\quad q=\frac{23.20}{46.4}=\boxed{\tfrac12}.\]

Check the candidate equilibrium at \(q=1/2\). The other owner permits with probability 0.9 in good weather and 0.6 in bad weather. An uninformed owner earns $60 by permitting, versus $0 by vetoing. For an informed owner, the signal-weighted payoff from permitting after a good report is \(U_+(1/2)=76.80>0\); after a bad report it is \(U_-(1/2)=-16.80<0\). Thus permitting after good and vetoing after bad are individually optimal, not just assumed. Buying yields \(76.80-16.80=\$60\) after its report cost, exactly the $60 from not buying, so randomising 50:50 is also optimal. This verifies a symmetric Nash equilibrium (not a claim that all equilibria have this form).

Finally, the optimal gross value of information, before subtracting the report price, is \(V(q)=\max\{0,U_+(q)\}+\max\{0,U_-(q)\}-U_0(q)=\max\{0,40-46.4q\}\). For \(0\le q<25/29\), a bad report is vetoed and the benefit is \(40-46.4q\); at \(q=25/29\) the bad-report voter is indifferent. For \(q>25/29\), even a bad report favors permitting, so the owner ignores its content and receives zero gross information benefit rather than a negative value. The information benefit therefore decreases then remains zero as the other owner is more likely to veto on bad weather. At the equilibrium \(q=1/2\), its gross benefit is $16.80, exactly the report price.

Assumptions

  • Each owner receives the stated personal trip payoff; cancellation produces zero trip payoff, excluding any report fee already paid.

  • Report accuracy is 80% in each weather state; purchased reports are conditionally independent given weather.

  • Acquisition choices, reports and simultaneous votes are private until the departure decision is determined.

  • All agents maximise their own expected monetary net payoff and know the stated common prior and rules.

Final Answer

A symmetric mixed equilibrium has \(q=\frac12\). Buy a report with probability 1/2; if bought, permit on 'good' and veto on 'bad'; without a report, permit. Each information option gives $60 expected net payoff against this policy.

Game theory and strategic decisions
↗

A wholesale produce dealer is still offering to buy one crate for £100 when a quality inspection is published. There is a 50% chance the result is bad, in which case the crate's fair value instantly falls to £90; otherwise it is worth £100. The dealer's own screen receives the result after a delay D, uniformly distributed between 0 and 1 minute. Independently, the dealer will withdraw the offer after half a minute even if the screen has not updated.

Supplier inquiries arrive at a Poisson rate of 4 per minute. Each is independently among the one-quarter of suppliers who learn the result immediately. If the result is bad, every such early-informed supplier will sell at the dealer's stale £100 bid when able; all other inquiries cause no trade. The first sale fills the dealer's single-crate order and closes it. Good news creates no stale-quote trades.

Assume the inspection result, screen delay and supplier-arrival process are mutually independent; the informed-status marks are independent of these and of one another.

What is the expected loss from this stale bid per inspection? The dealer separately earns £s of ordinary spread revenue per inspection cycle; what minimum s covers this expected loss?

Hint 1

Thin the supplier Poisson process to the suppliers who see the news early. Then find how long the quote can be hit.

Conditional on a bad result and delay \(D=d\), the exposure window is \(\operatorname{min}(d,\frac{1}{2})\), and the chance of at least one informed order is \(1-\operatorname{exp}(-\operatorname{min}(d,\frac{1}{2}))\). Average over D.

Only bad news creates an informed sale at a disadvantage. Each inquiry independently comes from an early-informed supplier with probability 1/4; thinning the rate-4 Poisson process gives an informed inquiry process of rate 1 per minute. Its first arrival time T is exponential with survival \(P(T>t)=e^{-t}\). Let \(W=\operatorname{min}(D,\frac{1}{2})\) be the time during which the stale quote remains available. The bid is hit exactly when \(T<W\). Conditional on \(D=d\) and bad news, \(P(\text{hit}\mid D=d,\text{bad})=1-e^{-\operatorname{min}(d,1/2)}\).

Because D is uniform on \([0,1]\), the conditional hit probability is

\[\begin{aligned}&\int_0^{1/2}(1-e^{-d})\,dd+\int_{1/2}^{1}(1-e^{-1/2})\,dd\\&=[1/2-(1-e^{-1/2})]+(1/2)(1-e^{-1/2})\\&=(1/2)e^{-1/2}\approx0.3032653.\end{aligned}\]

One stale purchase loses \(\text{£}100-\text{£}90=\text{£}10\). The bad result occurs with probability 1/2 independently of order arrivals and delay, hence \(\mathbb{E}[\text{loss}]=\frac{1}{2}\times\text{£}10\times\frac{1}{2}e^{-1/2}=\text{£}\frac{5}{2}e^{-1/2}\approx\text{£}1.5163266\). The expected net profit from this cycle is \(s-\frac{5}{2}e^{-1/2}\) pounds; break-even requires \(s\ge\text{£}1.5163266\). This s is an independent ordinary-trading revenue assumption, not a claim that every adverse fill earns a spread.

For an independent timing check, the first informed order arrives at time t with density \(e^{-t}\). It executes only for \(t<\frac{1}{2}\) and if \(D>t\), which for the uniform delay has probability \(1-t\). Thus \(P(\text{hit}\mid\text{bad})=\int_0^{1/2}e^{-t}(1-t)\,dt=[t e^{-t}]_0^{1/2}=(1/2)e^{-1/2}\), matching the conditioning calculation. Equality at a cancellation instant has probability zero.

Assumptions

  • Bad-news probability, publication lag and supplier arrival process are mutually independent.

  • Informed status of each inquiry is an independent \(\operatorname{Bernoulli}(\frac{1}{4})\) mark, and only bad-news informed suppliers execute against the quote.

  • The order has capacity one and is cancelled at the earlier of the screen update and the fixed half-minute deadline.

  • The ordinary £s spread revenue is independently earned per inspection cycle, not per adverse fill.

Final Answer

Expected adverse-selection loss = \(\text{£}\frac{5}{2}e^{-1/2}\approx\text{£}1.51633\) per cycle; minimum ordinary spread revenue s is that amount.

Market making and microstructure
↗

A concert venue can choose between two insurance arrangements for cancellation losses. Policy A leaves the venue with a $50 loss half the time and a $150 loss half the time. Policy B leaves it with either $0 or $200 in losses, also with equal probabilities. The costs of purchasing the policies are identical, and the distributions already include all cancellation-related payments. The venue starts with $1,000 before these losses.

Both policies have an expected loss of $100. Can you say which policy every decision-maker with increasing, concave utility of final wealth weakly prefers, without assuming a particular utility formula? Prove your answer using stop-loss expectations or an equivalent mean-preserving-spread argument. When is the preference strict?

Hint 1

Equal means alone are not enough. Compare the expected amount by which each loss exceeds an arbitrary threshold t.

The stop-loss functions are \(\frac{1}{2}(50-t)_{+}+\frac{1}{2}(150-t)_{+}\) for A and \(\frac{1}{2}(0-t)_{+}+\frac{1}{2}(200-t)_{+}\) for B. Check their difference on intervals with breakpoints at 0, 50, 150 and 200.

Let \(L_{A}\) and \(L_{B}\) denote the net losses. Their common expectation is $100. For any threshold t define the stop-loss function \(S_{L}(t)=\mathbb{E}[(L-t)_{+}]\), where \(x_{+}=\max(x,0)\). The proposed distributions give \(S_{A}(t)=\frac{1}{2}(50-t)_{+}+\frac{1}{2}(150-t)_{+}\) and \(S_{B}(t)=\frac{1}{2}(-t)_{+}+\frac{1}{2}(200-t)_{+}\).

For \(t\le0\), both equal \(100-t\), so \(S_{B}-S_{A}=0\). On \(0<t<50\), \(S_{A}=100-t\) and \(S_{B}=100-\frac{t}{2}\), yielding \(S_{B}-S_{A}=\frac{t}{2}>0\). On \(50\le t<150\), \(S_{A}=75-\frac{t}{2}\) and \(S_{B}=100-\frac{t}{2}\), so the difference is 25. On \(150\le t<200\), \(S_{A}=0\) and \(S_{B}=100-\frac{t}{2}\ge0\). For \(t\ge200\), both are zero. Consequently \(S_{A}(t)\le S_{B}(t)\) at every threshold, with equal mean; equivalently \(L_{A}\) is below \(L_{B}\) in convex order.

For a direct proof, couple the two distributions as follows: let \(P(L_{B}=200\mid L_{A}=50)=\frac{1}{4}\) and \(P(L_{B}=200\mid L_{A}=150)=\frac{3}{4}\). Since A has equal probabilities on 50 and 150, this produces \(P(L_{B}=200)=\frac{1}{2}\) and \(\mathbb{E}[L_{B}\mid L_{A}]=L_{A}\). With any concave utility u of wealth, conditional Jensen gives \(\mathbb{E}[u(1000-L_{B})\mid L_{A}]\le u(1000-\mathbb{E}[L_{B}\mid L_{A}])=u(1000-L_{A})\). Averaging shows \(\mathbb{E}[u(1000-L_{A})]\ge\mathbb{E}[u(1000-L_{B})]\). The inequality is strict for strictly concave u because the conditional distribution of \(L_{B}\) is nondegenerate for both A values. Linear (risk-neutral) utility is indifferent. A decision-maker whose utility is not concave need not share the ranking.

Assumptions

  • All payouts and premiums are already incorporated into the stated dollar loss distributions.

  • The comparison is in expected utility of final wealth, with no other uncertain wealth components.

  • Increasing concave utilities are defined over the possible final wealth interval [$800,$1,000].

Final Answer

Policy A is weakly preferred by every increasing concave final-wealth utility; it is strictly preferred for strictly concave utility. B is a mean-preserving spread of A.

Risk, covariance and diversification
8

You have a 1kg watermelon that is 99%…

Unlock the rest

Keep practising with 93 more questions, available hints and full worked solutions.

Explore Premium

© 2026 EverythingQuant. All Rights Reserved.