The Voleon Group Guide 2026: AUM and Careers

What is The Voleon Group?

Michael Kharitonov and Jon McAuliffe founded The Voleon Group in Berkeley in 2007, after quantitative work at D. E. Shaw and a stretch building production software. Machine learning is not a recent marketing layer. Voleon says flexible statistical models are how it approaches financial prediction, with scalability and risk management built into the research process.

Voleon Capital Management LP reported $29.38bn of regulatory assets under management in its 2026 Form ADV, and 158 employees at the registered adviser. That headcount is the regulated entity's figure, not automatically the whole group (Voleon Form ADV, 2026).

Voleon at a glance

• Founded: 2007
• Co-founders: Michael Kharitonov and Jon McAuliffe
• Headquarters: Berkeley, California
• Regulatory AUM: $29.38bn in the 2026 Form ADV
• Adviser employees: 158 in the same filing
• Approach: quantitative investing using statistical machine learning
• Clients: pooled investment vehicles and a small number of sovereign or official institutions in the filing data
• Official application route: Voleon careers
• Quantt directory: The Voleon Group

The AUM figure is regulatory, not a promise about current net assets or capacity. It should be dated whenever quoted.

Who founded Voleon?

Michael Kharitonov is co-founder and chief executive. His official biography says he developed quantitative equity and equity-derivatives strategies at D. E. Shaw before founding and leading technology company Netli, which Akamai acquired in 2007.

Jon McAuliffe is co-founder and co-chief investment officer. He previously researched and managed statistical-arbitrage strategies at D. E. Shaw, built recommendation and prediction systems in technology companies, and is an adjunct professor of statistics at UC Berkeley. Voleon now also names Vasco Chatalbashev, who joined in 2009, as co-chief investment officer (Voleon management, accessed 8 September 2026).

That combination is more revealing than the generic phrase "academic culture". The leadership histories join three disciplines: statistical research, production software and portfolio management. A Voleon research result has to survive all three.

How much does Voleon manage?

Voleon's 2026 Form ADV reports $29,377,232,147 of discretionary regulatory AUM across 23 accounts. The filing attributes about $26.3bn to pooled investment vehicles and $3.1bn to sovereign wealth funds or foreign official institutions. All reported assets were discretionary (Voleon Form ADV, 2026).

Regulatory AUM is the best current primary filing measure available, but it is not identical to:

• the market value in a 13F, which covers only specified US-listed holdings;
• private-fund gross asset value;
• investor net asset value on a later date.

Voleon does not state on its public site that funds are closed to new investors. Capacity is a live commercial fact, not something a careers page can settle. Prospective investors need current offering documents. This guide is not an investment solicitation.

What does Voleon trade?

Voleon describes quantitative strategies using a variety of asset classes and financial instruments in its regulatory disclosures. Its leadership page specifically says Chatalbashev helped build models and systems for trading equities, and public 13F filings confirm a sizeable listed-equity footprint. Those facts do not justify reducing the whole firm to "medium-frequency global equities".

The firm's own explanation focuses on the prediction problem: finding patterns in large, noisy datasets with flexible statistical models, then scaling the result under risk constraints. A useful way to break that down is:

1. Representation: turn raw market and non-market data into variables available at the decision time.
2. Generalisation: distinguish repeatable structure from chance patterns in a low signal-to-noise setting.
3. Portfolio construction: combine correlated forecasts under exposure, liquidity and cost limits.
4. Execution: trade without surrendering the forecast to impact and delay.
5. Monitoring: identify data failure and model drift without reacting to every random loss.

This is where Voleon differs from a company merely adding an ML model to a traditional research stack. Its public identity places statistical learning at the centre, while its management structure gives portfolio optimisation, market impact and software explicit senior ownership.

For foundations, read the machine learning for trading tutorial and statistical arbitrage guide. The machine learning in finance guide explains time-aware validation, including why random cross-validation introduces look-ahead bias.

Where does Voleon operate?

Voleon's official site identifies Berkeley as its headquarters, close to UC Berkeley. Regulatory filings use 1919 Shattuck Avenue in Berkeley as the principal office. Current and recent job adverts also show roles in London, New York and remote US locations, but job locations do not necessarily mean every city houses a full investment office.

Use the location on the live role, then ask how that team collaborates with Berkeley. Recent hiring has included London, New York and remote US roles. A posting in another city means Voleon recruits there. It does not mean every research function is replicated there (Voleon careers, accessed 8 September 2026).

What roles does Voleon recruit for?

Voleon's public materials seek people in statistics, computer science and business. Recent adverts have included research, machine-learning engineering, data, software, technical management, infrastructure, compliance and other operating roles. Vacancies are episodic. The Lever board showed no open postings when checked on 8 September 2026.

Three career tracks are especially relevant to quant candidates:

Research

Research roles can span prediction, portfolio optimisation and market-impact estimation. A doctorate is common in Voleon's leadership and public positioning, but the firm does not state that every researcher must hold one or come from a named university.

Machine-learning and research engineering

These roles sit between models and reliable systems. The hard problems include scalable training, reproducible features, efficient experimentation and turning research code into monitored production services.

Software and systems

Voleon explicitly presents scalability as part of the investment approach. Software roles are not merely back-office support: compute, data and deployment constraints determine which research ideas can be tested and traded.

Check the current advert for languages. There is no first-party list that makes Python, C++ or SQL compulsory across every quantitative seat.

How should you prepare for a Voleon interview?

Voleon does not publish a standard five-stage process, a mandatory take-home, topic weightings or a policy on expiring offers. Candidate anecdotes are clues, not the specification.

Preparation can still be firm-specific:

• Know one piece of work in depth. Explain the baseline, loss function, validation design and failure modes.
• Respect time. Show how features, labels and folds avoid future information and overlapping leakage.
• Prefer evidence to novelty. Explain when a linear or tree model is a better choice than a larger neural model.
• Discuss scale honestly. Cover compute, memory, reproducibility and what happens when the dataset or asset universe grows.
• Connect prediction to trading. A small forecast can be useless after turnover, impact and correlated exposure.

A good practice question is not "Which regulariser does Voleon ask about?" It is: "How would your model choice change if predictors are correlated, regimes drift and the final portfolio has a turnover constraint?" That joins statistical and investment reasoning without pretending to quote a real interview.

Use the quant research interview questions guide for drills. When applications reopen, tailor everything to the listed team.

What is a useful Voleon-specific question?

Ask how a team decides whether added model complexity has earned its place. The answer should connect out-of-sample improvement to stability, trading costs, compute and operational risk. That is more revealing than asking which fashionable architecture the firm uses.

For engineering roles, ask how researchers reproduce a dataset and model months later. For research roles, ask when portfolio constraints enter model evaluation. Both questions follow directly from Voleon's stated emphasis on scalability and risk management.

Compensation, culture and evidence limits

Voleon does not publish firm-wide total-pay ranges. Some adverts may carry location-specific base ranges. Treat those as role-and-date figures, not a house ladder. The US quant salary guide can frame questions. It cannot predict an offer.

Voleon says it values intellectual curiosity, flexibility, teamwork and open thinking. Those are first-party cultural aims, not proof of every employee's week. Dress codes, meeting-room folklore, annual hiring numbers and typical project length are not published with enough authority to repeat.

Frequently Asked Questions

How much does Voleon manage?

Voleon Capital Management reported $29.38bn of regulatory AUM in its 2026 Form ADV. Because assets and filing measures change, quote the date and do not substitute the 13F equity total.

Who founded The Voleon Group?

Michael Kharitonov and Jon McAuliffe co-founded Voleon in 2007. Both had quantitative-research experience at D. E. Shaw; Kharitonov is CEO and McAuliffe is co-CIO.

Does Voleon only trade equities?

No public source supports "only". Equities are clearly important, but the regulatory description permits quantitative strategies across a variety of asset classes and instruments.

Do you need a PhD to work at Voleon?

Not for every role. Many leaders and researchers hold doctorates, and research jobs may require advanced training, but software, infrastructure, operations and business roles have different criteria.

Is Voleon hiring now?

Its public job board had no open postings when checked on 8 September 2026. Openings change, so use the official board rather than relying on an indexed or expired advert.