Capital Fund Management (CFM): Research, Jobs & Career Preparation

Capital Fund Management (CFM) is a quantitative investment manager based in Paris. Its official website describes a global business managing alternative investment strategies for institutional clients and wealth managers. The company history dates its establishment to 1991.

The contact page lists Paris, New York, London, Toronto and Shanghai. An office listing does not establish that every job family hires there. Select the location on the individual vacancy; the examples below include Paris research and technology descriptions and a London risk role.

The homepage reports $33bn+ AUM, labeled August 2026. This employer-reported figure does not establish salary levels or hiring plans. The checked official sources do not establish a consistent current company-wide headcount.

Checked 3 October 2026. Job descriptions are a dated research snapshot. Follow the live jobs portal before applying, since an accessible description can outlast a vacancy’s active recruitment period.

CFM’s approach page connects four activities: research into statistical patterns, technology that implements and monitors strategies, risk management, and execution research. This makes “research at CFM” broader than choosing a predictive model. Portfolio decisions and the costs of implementing them also matter.

The official strategy descriptions show several distinct programs:

These investment objectives do not establish realized returns, proprietary signals or a universal trading horizon.

For a career decision, ask where you would contribute: investigating data, designing allocations, improving training, implementing production components or assessing exposures. Read the vacancy for that answer.

Each link leads to an employer description reviewed on 3 October 2026. These are selected examples, not an exhaustive vacancy list. Requirements are attached to that description; they should not be transferred to another CFM role.

For the ML researcher, the preferred additions include inference optimization, orchestration, alignment methods and CUDA. They are separated from the core requirements in the source. Treating every preferred item as compulsory would misread the vacancy.

The junior role accepts experience in several related areas; economics research is more specialized. Read the complete posting before assuming equivalence.

Choose the direction by the question you can answer convincingly with your own work. The following distinctions are our interpretation for preparation, rather than CFM’s assessment rubric.

These paths overlap: researchers explain implementation decisions, and engineers understand outputs. Use the distinction to allocate preparation time.

Do you need a PhD? The economics requirement, the ML preference and the junior developer’s alternative degrees differ. A preference signals expectations without establishing a universal entry rule.

Verify language, experience and location before applying. A web page’s language does not establish the required working language.

This independent learning exercise turns a vague interest in systematic investing into evidence you can explain. It is not a CFM take-home test. Start small enough to finish and audit.

Use a permitted public time series or clearly labeled synthetic data. Ask whether a lagged variable improves prediction over a simple baseline. Define decision time, prediction horizon and a useful result before examining the final test period.

1. Make the observations auditable. Record the provider, download date, transformations and missing-data policy. For economic data, distinguish the period a number describes from when it became available; revised data can answer a different question from data available at the time.
2. Define a baseline. Compare the model with a simple alternative on the same samples. Explain what evidence justifies additional complexity.
3. Separate time periods. Use an earlier interval for development and a later interval for the final evaluation. Fit transformations on the development interval. If you repeatedly inspect the final interval and change the rule, it is no longer an untouched test.
4. Show failure cases. Report deterioration, influential observations and how many variants you tried. Explain uncertainty limitations with dependent errors or overlapping return windows.
5. Keep prediction and trading separate. If you add a hypothetical portfolio, explain when positions change, how exposures are limited and how turnover produces costs. Better prediction need not translate into better net returns.

For a simple cost illustration, suppose a modeled one-way turnover of 0.40 incurs an assumed 5 basis points per unit of traded notional. The deduction is 2 basis points of portfolio value. A gross modeled return of 1 basis point becomes −1 basis point after that deduction. These numbers are invented; the exercise demonstrates accounting, not achievable trading performance.

Deliver a README, runnable analysis, data dictionary, calculation tests and results table. End with a decision: reject the idea, investigate a named uncertainty or run a specified experiment. Explain what would change your decision. A negative result with an audit trail is useful evidence.

Extend the project with an independent engineering exercise. Separate ingestion, features, evaluation and reporting. Finish a small working pipeline before optimizing it.

Test duplicate timestamps, missing observations, changed schemas and late corrections against known results. Preserve raw input and record transformations so you can explain any change on rerunning.

For a portfolio simulation, test the accounting on a tiny hand-calculated example. Check that a price change produces the expected P&L, that changing a position creates turnover, and that costs are deducted once. Test an empty portfolio and a constraint violation. A plausible cumulative-return chart cannot replace these checks.

Measure runtime and memory on a fixed workload and record the environment. Change one choice at a time and compare correctness with the baseline. For ML, begin with a modest reproducible model; claim distributed-training experience only if you have run and diagnosed it.

Prepare an incident note: failure, observation, diagnosis, fix and regression check. Explain the pipeline’s limitations alongside its measured behavior.

Start with Our approach and the particular role you want. Then browse Our thinking, which separates public commentary and research material. Choose one item relevant to the role rather than collecting titles for a reading list.

Make a reading note covering question, evidence, assumptions and a limitation. Follow links to full articles or papers before discussing the method. Public writing does not disclose every live model.

Use the following suggested sequence, adjusting it to your starting point:

The QuantSteps research path organizes general foundations; the assessment helps identify gaps. These independent resources are not a CFM curriculum or a hiring-outcome prediction.

CFM’s careers page includes students and interns, and the linked data-quality internship provides a concrete example. Its student level and language requirements belong to that opportunity. They do not establish an annual research-internship intake, a shared deadline or guaranteed progression into a full-time job.

Verify internship enrollment eligibility, length, start date, location and any university arrangement. The data-quality description does not state a complete timetable; confirm through the application process.

Before applying:

• Save the exact role title and link, then distinguish required qualifications from preferred additions.
• Map three responsibilities to specific evidence from your work. State your personal contribution to collaborative projects.
• Make the résumé’s project descriptions explain the question, method and result, with a limitation when it matters.
• Check that any linked artifact can be opened and reproduced without private credentials or proprietary data.
• Confirm location, language and work-authorization expectations with the hiring team when the process allows.

A focused application should answer what you can independently investigate, build or explain.

The checked descriptions establish no fixed interview sequence or verified question bank. As preparation, explain a hypothesis and its tests, code you wrote, an error you found and a decision’s trade-offs. These suggestions do not establish what CFM will ask.

Ask the recruiter about format, duration, permitted tools and whether a research presentation is expected.

No numeric compensation range was disclosed in the six descriptions summarized here. This guide therefore does not present a CFM salary estimate as an employer-confirmed fact. Paris and London, an internship and a senior role, and base salary and total compensation need separate comparisons.

When compensation is discussed, request the currency, annual base, bonus terms, any first-year guarantee, deferrals and conditions attached to one-time payments. Keep discretionary bonus expectations separate from guaranteed amounts.

Ask who reviews research, how models enter production, how failures are diagnosed and what success means in the first months. Use the answers to assess fit.