For tomorrow’s quants, Python is essential; AI isn’t

Proportion of PhDs in quant teams is sliding, as employers focus on all-round skills

Key points Summary Q&A

• The Tomorrow’s Quants series explores the skills needed by new quant recruits, drawing on a survey of 39 employers, and six podcast interviews with master’s programme directors.
• Employers prefer candidates with coding skills and practical knowledge over advanced mathematical prowess, with only four in 10 recent hires now holding a PhD.
• While data science skills are valued, AI-specific knowledge is not. Most quant employers do not expect new hires to spend much time on AI-related projects.
• Among recent hires, 42% have a maths background, while roughly one-third have a finance, engineering, or statistics specialisation.

These summaries are generated by an LLM, but edited by humans.

The Tomorrow’s Quants series examines the skills and traits required of new quant recruits via a survey of major employers – and how universities are preparing them for the workplace, via a series of podcast interviews. The next article in this series will look at soft skills and employers’ hiring plans.

If you’re reading this article, then you’re used to being tested, so please excuse this simple, one-question quiz. The quote below describes the desired profile for a new hire at a large, financial services firm. But for what role?

“You just need to be quick on your feet, good at coding, know enough about finance, and then learn on the job.”

Maybe it’s some kind of technology position? A junior at one of the consulting firms? Or an investment bank’s analyst programme?

Admittedly, the wider context is a bit of a giveaway.

But congratulations, nevertheless, if you realised the quote describes a good quant candidate. The speaker is the global head of the quant group at a big US bank, who was one of 39 participants in our Tomorrow’s Quants survey – an attempt to shed light on the qualities that employers are looking for in new recruits, and what that means for the schools that are preparing those recruits for the workplace.

Related

Tomorrow’s Quants: what it takes to be a next-gen modeller 

Employers increasingly prize mix of hard and soft skills, Risk.net survey reveals

One of the takeaways is that employers generally aren’t looking for the world’s most promising mathematicians. Being a stochastic whizz certainly won’t hurt a candidate’s chances, but it’s not a necessary – or sufficient – condition on its own. Employers want people who can build, validate, mend and generally support the business in a wide range of maths-y tasks.

You can see it in the quote from that US bank’s quant head, or in the finding that roughly four in 10 junior quants now hold a PhD – a decade ago, the quant head says it would have been more like nine in 10 at his bank. For employers in the survey, nearly eight in 10 junior quants have instead arrived via master’s programmes.

This reflects two changes. First, the work being done by many quants has become less hardcore – requiring strong maths skills, but not necessarily cutting-edge ones. And second, master’s courses are preparing more workplace-ready candidates.

The data – and, again, that quote – shows that these candidates need to be strong coders. Master’s programmes have adapted in recent years to ensure coding skills are a core part of the curriculum. Clearly, they’ve got their priorities right on that score.

But in recent years, a growing focus of many programmes has been to emphasise a variety of skills relating to machine learning and artificial intelligence. The Tomorrow’s Quant data suggests employers see data science as a useful part of the toolkit for a new hire – but they’re not currently expecting junior quants to spend much of their time working on AI-related projects. Are academic institutions devoting too much time to the topic or are employers not devoting enough?

Language students

In one of the key sections of the survey, respondents were asked to say how important it is for a new hire to have proficiency in each of 10 areas, rating them from 1 to 5, or unnecessary to essential (see figure below). Results are shown for all respondents, and also broken down into six cohorts – representing senior quants who manage specific teams. It should be noted that some of these cohorts are small – only three respondents run algo trading teams, for example, and only five lead model validation teams – but their responses are often sufficiently consistent with each other, and distinct from other cohorts, that it’s interesting to consider them as separate cases.

Across the population as a whole, one skill emerges as critical: the ability to code in at least one language. This scores an average of 4.8 across our respondents. Some also want candidates to have coding skills in multiple languages, but at 3.5, it is an expectation that is not as universally shared.

 

If you’re picking one language to specialise in, then it should probably be Python (see figure below). This programming language, which gained prominence in finance in the early 2010s, is seen as the most desirable option across the respondent group as a whole. Among the pricing and modelling cohort, Python is nudged into a close second by C++.

Other coding languages that historically had significance in finance, such as Java, VBA, R and Matlab, now seem to have slipped in prominence.

 

The next most popular skill sets are those classically associated with quant careers in banking – knowledge of specific products or asset classes, and derivatives pricing models – but not all cohorts value them equally. Employers who lead algo trading or quant investing teams care much less about a candidate’s proficiency with Black-Scholes.

One option missing from the list of 10 was a popular write-in choice, with eight of 16 respondents insisting that new hires should have a solid knowledge of statistics and probability theory – all of them scored it a 5.

For financial institutions hiring entry-level quants, evaluating this bundle of foundational maths-stats-and-finance competencies is often a primary step in the recruitment process.

Vladimir Lucic, head quant at Marex Solutions, built his team from the ground up when the firm expanded into exotic derivatives and market-making. “I have a baseline competence test of old school mathematics that [candidates] have to know if they want to become a quant,” he says. “There’s a fundamental basis of finance that we think is absolutely necessary, no matter what other tools or skills people bring to the team”.

There are many areas where the current state of the art in AI is either inapplicable or not computationally feasible

Head of algo trading quants at a major US bank

He explains that everybody in his team, senior or junior, will eventually end up reading sophisticated research papers, and they need to understand the mathematics in those papers well enough to be able to reproduce and test them.

These criteria led Lucic to hire several of his team members from École Polytechnique in Paris, which is known for providing graduates with a strong theoretical backbone, but also from Imperial College London, where he teaches, and Imperial College Business School.

At the other end of the scale from these must-have skills is knowledge of deep learning and neural network theory. Perhaps surprisingly, across the population, this subject tied for last place alongside portfolio theory, and enthusiasm was tepid even within the cohorts – quants leading algo trading teams rated it their fourth most important skill set, while it was fifth for front-office strats, and sixth among quant investors. Despite ongoing testing and exploration of these techniques in finance over the past decade, they appear to lie outside the skill set demanded of early career quants. For now, at least.

In contrast, data science, though closely tied to neural networks, scored 3.5 across all respondents – putting it joint fourth – and was rated significantly higher by quants leading algo trading, front-office strat and quant investing teams, where data-driven approaches were a core competency long before the current AI bandwagon started rolling. Data science skills were least prized by model validation and pricing teams.

No displacement?

A thought experiment in the survey reveals the extent of AI’s current import for the different cohorts. Respondents were asked to estimate the increase in team size that would be necessary to maintain current productivity levels, if AI were suddenly unavailable. About 70% of participants said no significant increase in staff would be needed. Or, to put it another way, at the vast majority of these firms, AI is not yet displacing any human hires.

There were exceptions. Roughly one-quarter of employers estimated they would need to increase staff by up to 30% to make up for lost AI production. And one hedge fund manager estimated staffing would need to grow by 30–70%.

That challenges the perception, held by some outsiders, that AI already has a central role in teams of pricing or risk management quants. Some employers note that candidates are among those making this mistake – with some apparently coming to interviews prepared to showcase their knowledge on AI-related topics, and being surprised to receive a grilling on finance and maths instead.

“There are many areas where the current state of the art in AI is either inapplicable or not computationally feasible,” notes the head of algo trading quants at a major US bank.

The main risk I see with AI in quantitative finance is training a generation of new quants to get used to pre-canned, turn-the-crank, black box solutions, which hinders true creativity

Carlo Acerbi, Adia and Risknowledge

Despite some notable successes on the sell side – deep hedging being one of the best-known examples – financial applications are still limited.

Unlike traditional AI architectures, generative AI (GenAI) is making a discernible – but general – impact on productivity across various desks. Applications often do not pertain to the core business of the institutions. The most significant current application appears to be assisting developers with coding, a practice that has now become standard in the industry. This could be a double-edged sword for early-career quants – giving them an assistant to help crunch through coding tasks, but also potentially reducing demand for junior quant developers.

Some do not see the status quo changing any time soon. When asked what quant finance-related tasks they expected to be performed by AI within the next 12 months – an open question – many respondents volunteered fairly generic applications (“accelerates daily tasks”, “document summarisation”, “report generation”) or dismissed it (“no specific task”, “none”, “none in my area”).

Related

Quantcast Master’s Series: Dan Stefanica and Jim Gatheral 

Baruch College leaders on how they manage the top-ranked quant finance master’s programme

Others see it becoming embedded in the coding workflow. And a handful expect it to start encroaching on quant-specific tasks (“factor creation, model creation”, “model description and prototyping”, “optimisation of trading strategies”).

Hamza Bahaji, head of financial engineering and investment solutions with Amundi ETF, explains why there has – so far – been little impact on the desired profile for new recruits: “We use AI for coding and for data generation. We need to hire a lot of people for that.” But, he says, “we need ALM experts with deep knowledge of the asset classes we cover … with a double expertise in finance and AI”, which isn’t the profile of a starter.

The bigger questions are about how AI will affect ongoing demand for quants, and their development. There are some concerns here (see figure below).

“The main risk I see with AI in quantitative finance is training a generation of new quants to get used to pre-canned, turn-the-crank, black box solutions, which hinders true creativity, deep subject-matter understanding and critical thinking”, says Carlo Acerbi, quantitative research and development lead at Adia – the Abu Dhabi Investment Authority – and head of research and development at Risknowledge, a risk management consultancy firm.

 

Nonetheless, a substantial number of respondents express optimism regarding the future impact of GenAI on the professional development of young quants. While there are concerns about the potential outsourcing of knowledge and the challenges of educating and evaluating students, many believe these obstacles can be addressed. Given that AI technology is poised for continued evolution, almost half of the respondents are confident in its ability to effectively support the career development of junior quants. Ultimately, the majority view the expanding role of AI in quant finance favourably, believing the pros outweigh the cons.

“We need industry wide-education of quants on how to use AI tools, spanning coding, data analysis and even prompt engineering,” chimes in a quant analytics managing director at a major European bank.

Mastering the workplace

Educators play a crucial role in shaping knowledge and managing expectations to align career paths with job requirements. Nearly all junior quants hired in the past two years – based on our respondents’ data – possess a master’s degree.

PhD-holders comprise about 40% of the new quant workforce at the surveyed institutions. A number that for some of them is lower than it was in the past.

“We used to have 90% of PhDs 10–15 years ago. Now, it is much less. Partly, that is because graduates coming out of master’s programmes are prepared better than they used to be,” says the US bank’s head quant, who was quoted at the top of this article.

Related

Quant Finance Master’s Guide 2025 

Risk.net’s guide to the world’s leading quant master’s programmes, with the top 25 schools ranked

But it’s also because a good chunk of the job in a big quant team has gone from writing pricing models to countless other tasks for which PhD-level quants are not necessary, such as regulation-related assessments and reporting – hence the comment about being “quick on your feet”, good at coding, and familiar enough with finance to pick up the rest.

“That’s where quant finance master’s programmes are good at preparing people,” he adds.

The backgrounds for recent hires vary considerably. Across the respondents, only one-third of junior quants hired in the past two years have an academic background in finance – a figure that is roughly consistent across all responding cohorts. A larger proportion, about 42%, hail from mathematics backgrounds. This is particularly true for algo trading roles, where they represent two-thirds of juniors, yet rarer in risk management roles. Approximately one-third of juniors have backgrounds in engineering or statistics, and about one-quarter in physics or data science. It’s important to note that many candidates possess multiple specialisations, which explains why the total exceeds 100%.

A final takeaway is that one of the most compelling arguments candidates have to demonstrate they deserve to become quant researchers, is to show they have already produced quantitative research.

 

Marex’s Lucic says that having published in a peer-reviewed journal is a huge advantage for candidates. Not only does it demonstrate technical capability, but knowing that a research paper has passed a review process gives him added confidence when evaluating the candidate’s work.

Editing by Duncan Wood