What Does a Quant Do? Roles, Daily Work and a Worked Example

A quant uses mathematics, statistics and software to answer financial questions that affect investment, trading, valuation or risk. A researcher might ask whether information available today helps predict a later return. A trader might decide how much risk to take and how to execute it. A developer might make the data and trading system reliable. A bank pricing quant might estimate a derivative’s value and its sensitivity to market changes.

The shared workflow is define the question, inspect the inputs, implement an approach, test its limits and communicate a decision. The balance changes with the job. There is no single “quant day” or universal stack.

The role descriptions below draw on employer pages checked on 3 October 2026. The worked scenarios are explicitly invented teaching examples, not employee diaries or accounts of proprietary strategies.

Quant researcher (QR)

A QR investigates a question with data and experiments. For investment research, that may mean defining a target, checking when each input becomes observable, constructing features and comparing a candidate model with a baseline. The researcher must explain what survives validation and what fails. Two Sigma describes systematic hypothesis testing as part of its research work. Two Sigma research careers.

The job can extend into implementation and live-model analysis. Jane Street’s researcher description includes experiment design, time series, feature engineering and studying production trading behaviour. A notebook is a means of investigation; a defensible result is the deliverable. Jane Street researcher role.

Quant trader (QT)

A QT makes quantitative trading decisions and manages risk. Jane Street’s description includes signals, strategy execution, models, statistical analysis and algorithmic systems, showing that trading and research duties can overlap. Jane Street trader role.

Quant developer (QD)

A QD or engineer builds software used by research, pricing or trading teams. Illustrative tasks include a reproducible data pipeline, a tested numerical library or an execution service with monitoring and recovery. The title can describe different engineering specialties, so ask whether the team needs numerical methods, data engineering or performance-sensitive systems. Jane Street calls its engineers closely involved throughout the business and stresses programming and collaboration. Jane Street software engineering role.

Bank pricing quant or desk strategist

Pricing work asks what a transaction is worth under stated assumptions, and how its value changes when market inputs change. A teaching example is implementing an option valuation method, checking it against a known benchmark and investigating unstable sensitivities. Goldman Sachs describes quantitative strategists working on pricing, prediction, portfolio allocation and trading automation; banking quant work is not confined to one model type. Goldman Sachs engineering careers.

Risk quant and model validation quant

A risk quant builds and monitors models of exposures or losses. Bank of America’s market-risk role includes market-data verification, benchmarking, backtesting, stress testing, documentation and explaining results to stakeholders. Bank of America market-risk role.

A validator examines whether another model is reliable for its intended use. That includes assumptions, construction, outcomes and limitations. Goldman Sachs describes model risk management as independent controls around model development, implementation and use. The Federal Reserve’s current guidance also distinguishes validation and monitoring from merely producing a model. Goldman Sachs risk careers, Federal Reserve model-risk guidance.

These illustrative deliverables make the differences easier to recognise. They are examples of work products, not a standard checklist used by every employer.

The handoff matters. A researcher must specify input availability and trading timing; an engineer needs an unambiguous specification; a trader needs to understand the model’s limits. One person can perform several functions. Jane Street describes overlapping disciplines; this table explains responsibilities rather than a rigid organisation chart. Jane Street trader role.

At an investment research team, the central question may be whether a proposed insight improves an investment process. Two Sigma emphasises patterns in data and systematic hypotheses. Ask which horizon, asset class and portfolio decision your research serves. A daily equity signal and a longer-horizon allocation study need different tests. Two Sigma research careers.

At a trading firm, research and execution can be closely connected. Jane Street’s researcher description includes analysing how models trade in production. Ask how much of the role involves exploratory work, live strategy behaviour and systems performance. Jane Street researcher role.

At a bank, the label can cover pricing and trading analytics, market-risk modelling or independent model controls. Goldman Sachs lists several of those functions across engineering and risk. The desk and business purpose tell you more than “bank quant” alone. Goldman Sachs engineering careers, Goldman Sachs risk careers.

These are examples of mandates, not a ranking of intellectual difficulty or a claim that every fund, prop firm or bank has the same culture.

Imagine a researcher investigating this deliberately simple hypothesis: an up interval predicts a gain that persists after a one-close execution delay. This is our constructed example. All 20 prices are synthetic; there are no market dates, dividends or employer data.

The researcher first records the sample and convention. Observations 1–12 form the development segment. Evaluation returns are the eight intervals ending at observations 13–20. The rule is fixed before examining that latter segment.

Check that prices are positive, ordered and complete. Compute each interval return as next price divided by previous price minus one. After observing a return at a close, schedule the target position for the following close, then earn the interval after that fill. At observation 2, for example, the up move from 100 to 101 sets a long target to be filled at observation 3; its first earned return is from 102 to 101.8. Start with two cash intervals.

The scheduled next-close fill is still an idealised execution assumption. Real research must examine order deadlines, the available fill price and slippage. The cost is an assumed fraction of wealth per unit position change, not an observed spread. One basis point (bp) is 0.01 percentage point: at 10 bps, each unit position change deducts 0.10% of wealth under this example’s convention. No liquidation cost is added at the end.

Here is the complete calculation. It uses only Python’s standard library and preserves the existing position at the segment boundary.

from math import prod

prices = [100, 101, 102, 101.8, 102.8, 103.8,
          103.6, 104.6, 105.6, 105.4, 106.4, 107.4,
          107.3, 107.4, 107.5, 107.4, 107.5, 107.6,
          107.5, 107.6]
returns = [b / a - 1 for a, b in zip(prices, prices[1:])]
positions = [0, 0] + [int(r > 0) for r in returns[:-2]]
changes = [abs(p - (positions[i - 1] if i else 0))
           for i, p in enumerate(positions)]

def result(start, stop, cost_bps):
    cost = cost_bps / 10_000
    return prod(1 + positions[i] * returns[i] - cost * changes[i]
                for i in range(start, stop)) - 1

for label, start, stop in [("Development", 0, 11),
                         ("Evaluation", 11, 19)]:
    print(label, [round(100 * result(start, stop, c), 4)
                  for c in (0, 10, 25)])
print("Evaluation buy-and-hold:",
      round(100 * (prices[-1] / prices[11] - 1), 4))

The output changes the decision:

Evaluation buy-and-hold returns +0.1862% before costs. The strategy changes position five times during evaluation. Its gross result already fails this comparator; the assumed costs make it worse. This does not establish that momentum fails generally. It establishes that this rule has not earned support from this constructed sample.

The researcher should not hide the later loss or repeatedly change the lookback until this same evaluation segment looks good. Once inspected, it is no longer untouched evidence for a new variant. Further work needs a recorded hypothesis and new validation data.

The research brief that goes to a reviewer

Question: does the fixed prior-up rule outperform a stated comparator? Data: 20 invented closes, development through observation 12. Method: delayed binary position, compounding and explicit turnover costs. Result: evaluation gross 0.0000%; net −0.4993% at 10 bps; comparator +0.1862% gross. Decision: no basis for deployment. Limitations: tiny constructed sample, idealised execution, no financing or dividend effects, no uncertainty estimate. Next test: use a longer appropriately sourced dataset, dated available inputs, executable-price assumptions and a fresh validation segment.

That brief demonstrates an essential part of the work: turning an appealing result into a decision with limits. In this example, the useful output is a transparent rejection.

These are also invented task sequences, rather than hour-by-hour employee schedules.

Trading example: a dashboard reports worse fills during a volatile session. First check whether the apparent deterioration comes from stale reference prices, a reporting error or actual execution. Review inventory, limits and the strategy’s permitted response. If intervention is authorised, record the reason and communicate it. Afterwards, compare execution against the stated benchmark and separate market movement from trading costs. The deliverable is a justified decision and review, not an explanation that every losing trade was a model failure.

Development example: a pricing service starts responding slowly after a change. Reproduce the issue, profile the relevant workload and identify the bottleneck. Change the implementation, compare values against the previous version, examine boundary inputs and measure the performance gain. Submit the change for review, deploy through the team’s process and keep a recovery plan. The deliverable is a verified change with operational evidence, not simply a faster function.

Both examples need judgment about what to investigate first, coordination with colleagues and a record that another person can inspect.

Choose skills by the work product you want to produce. For a research task, practise probability, statistical inference, regression, experiment design and readable analysis code. For an engineering task, practise data structures, testing, profiling and reliable systems. For pricing work, add the product’s financial assumptions and appropriate numerical methods. For risk or validation, practise interpreting scenarios, challenging assumptions and writing technical evidence.

The employer examples show why a universal language checklist is misleading. Jane Street’s researcher role explicitly asks for comfort with Python, while its trader role says a particular programming language is not required. Its software engineering page describes OCaml as a primary tool and Python in research and trading; it does not require incoming engineers to know those languages. Researcher role, trader role, engineering role.

Useful practice tools include a notebook or script for analysis, a database query for checking inputs and version control for tracking changes. They support the task; possessing each tool is not proof of competence. Being able to explain which data were available, reproduce a result and identify a weakness is stronger evidence than listing libraries.

Communication is part of the technical work. Bank of America’s market-risk description explicitly includes documentation and explaining analysis to stakeholders. Practise a brief that states the question, method, result and decision in language a colleague can challenge. Market-risk role.

There is no defensible single schedule for all quant roles. A position serving live markets may have market-session and handover obligations. A research deadline, model review or production incident can create a different demand. Those are factors to investigate, not evidence for a universal weekly-hour average.

Ask the team which time zones it covers, who monitors live systems, whether there is an on-call rota, how market events affect the day and what deadline periods look like. A scheduled-hours field in a job posting does not describe every employee’s observed workload.

Assess the source of pressure as well as time. Are you responsible for a live decision, an unreliable service, a research result that may fail, or a model challenge others disagree with? Your tolerance for those tasks is a better career signal than an unsourced “best work-life balance” ranking.

Start with a task whose output someone else can check. In the free Returns & risk lab, inspect the supplied 20 synthetic prices, calculate consecutive returns, build the compounded curve, identify the drawdown and write a short explanation. Export the completed project so the calculation and brief can be reviewed together.

The lab teaches arithmetic and research communication. It is not the signal backtest above, a general-purpose CSV cleaning service or an employer’s real assignment. Neither exercise establishes investment performance or job readiness by itself.

Notice which part you want to investigate further. If you enjoy questioning the evidence, explore the QR learning path. If you prefer implementation and reliability, inspect the QD path; if you prefer trading decisions and execution, inspect the QT path. The current substantial curriculum is research-focused; specialist engineering and trading lessons are planned. The assessment helps you choose a starting point for study, not certify suitability for an employer.