Reading the QuantMinds 2026 programme as a map of modern quantitative and computational finance
If one wanted to explain quantitative finance to an intelligent outsider twenty years ago, the answer would have been comparatively simple. Quants built mathematical models of financial markets. They calculated the prices of derivatives, measured risk and devised systematic trading strategies. Their natural habitat was somewhere between a mathematics department and a dealing room, populated by stochastic differential equations, Monte Carlo simulations and C++.
The programme for QuantMinds International 2026, taking place in London on 16–19 November, suggests that this description is no longer wrong so much as radically incomplete. The conference is divided into tracks on investment, derivatives, risk, artificial intelligence, computational infrastructure and, on its final day, quantum computing. Alongside them run workshops on machine learning and volatility.
Reading the programme as evidence rather than advertising reveals something interesting. Quantitative finance in 2026 is becoming less a particular branch of applied mathematics and more a science of decision-making under uncertainty implemented in software.
And almost every word in that definition matters.
Mathematics did not disappear when AI arrived
The easiest mistake would be to look at the programme, count the references to AI and conclude that traditional quantitative finance has been superseded by machine learning.
It has not.
There are still stochastic-volatility models, local-volatility models, volatility smiles, term structures, hedging problems and calibration problems. There is an entire workshop beginning with Black–Scholes and moving through stochastic, local, rough and path-dependent volatility. Another session considers deep learning for pricing and hedging, while stochastic-volatility modelling proceeds alongside it.
Later in the conference one encounters SABR, XVA, Bachelier models, Monte Carlo Greeks, arbitrage-free volatility surfaces and exotic derivatives. There are questions that would have been immediately recognisable to a derivatives quant in 1996.
What has changed is the computational machinery available for answering them.
A particularly revealing example in the programme is a neural-network implementation of the SABR interest-rate model trained on more than 200 million simulated points. The mathematical object is familiar; the method for interrogating it is new. Instead of repeatedly performing an expensive numerical calculation, one can calculate an enormous collection of examples and train a machine to approximate the resulting map.
This is an increasingly important pattern.
The old distinction was often between an analytical solution and a numerical one. The emerging distinction is more complicated:
mathematical model -> simulation -> data -> learned surrogate -> real-time decision.
Machine learning therefore does not necessarily replace the model. Sometimes it becomes a very sophisticated numerical method for operating the model.
The quant is becoming a computational scientist
One keynote description almost states the new worldview explicitly. Quantitative finance, it says, advanced through three things: better mathematics, better data and faster computers. AI now constitutes a fourth wave, but one dependent on a new architecture of CPUs, GPUs, memory, networking and software. The claimed applications range from signal discovery and portfolio construction to derivative pricing, risk management and agentic workflows.
That is an important shift in emphasis.
A model is no longer useful merely because it is mathematically elegant. One must ask how quickly it calibrates, how efficiently it runs on modern hardware, whether its gradients can be computed, whether it can be distributed, whether it can survive production workloads and whether thousands or millions of instances can be evaluated within a trading or risk system.
Hence the conference programme contains topics that could comfortably appear at a computer-science conference: GPU acceleration, cloud-native systems, microservices, graph databases, differentiable programming and formal methods. The pre-conference technology summit explicitly groups machine learning, quantum computing, GPU acceleration and cloud infrastructure into the modern quant technology stack.
“Computational finance” therefore means considerably more in 2026 than implementing a finite-difference scheme.
It includes the architecture of computation itself.
AI is moving from model to co-worker
There is another transition hidden in the programme.
The first generation of financial machine learning mostly asked familiar statistical questions with unfamiliar tools: can a neural network predict returns? Can it classify documents? Can it approximate an option price?
The 2026 programme repeatedly asks a different question:
Can the machine participate in the research process itself?
There are sessions on agentic AI for mathematical finance, LLMs producing trading signals, AI credit analysts and systems that process financial documents. But one of the most revealing descriptions concerns the “agent harness”.
The underlying language model is only one component. Around it sits software that maintains state, controls access to tools and data, checks intermediate results, imposes constraints and turns failures into further actions. Probabilistic AI is permitted to reason about ambiguous things, while calculations, permissions and validation remain explicit and inspectable.
That architecture tells us a great deal about where serious financial AI seems to be heading.
The interesting object is no longer simply:
a very clever model.
It is:
a model enclosed within a controlled computational system.
That resembles quantitative finance itself. A human trader is not allowed simply to have intuitions and move arbitrary amounts of money. Around the trader are limits, market data, pricing systems, risk systems, execution systems, compliance procedures and independent checks.
Agentic AI is acquiring something similar.
And finance remains suspicious of its machines
For all the enthusiasm, the programme is strikingly sceptical about AI.
There are discussions of whether AI is genuinely intelligent or simply powerful automation. There are sessions on assessing retrieval-augmented-generation systems using metrics, threshold calibration and monitoring. One roundtable asks whether AI represents the next stage of finance or merely a “risky parrot” capable of amplifying systemic risk.
Even the hackathon is revealing. Its stated 2025 problem, retained in the 2026 programme pending announcement of the new challenge, was not “make an AI trade better”. It was to construct prompts resistant to confirmation bias, framing effects, priming and anchoring, and then measure the remaining bias statistically.
That may be a preview of a major subdiscipline.
Twenty years ago model validation meant asking whether a mathematical pricing model had the right assumptions and behaved correctly.
Tomorrow’s model validator may have to ask:
What happens when the model interpreting my model is itself probabilistic?
The problem becomes recursively quantitative.
Synthetic markets are becoming laboratories
Finance suffers from an unusual scientific problem: we possess only one history.
There was only one 2008 financial crisis. Only one pandemic market shock. Only one path taken by today’s yield curve. Rare events are precisely the events risk managers most desperately need to understand, yet by definition there are few examples of them.
Generative modelling offers an intriguing response.
A session on YieldGAN, for example, proposes generating synthetic yield-curve scenarios exhibiting nonlinear dependencies, volatility clustering, curve-shape changes and regime-sensitive correlations. The stated intention is explicitly not to replace conventional models, but to test portfolios against plausible environments that have never actually occurred.
This suggests an increasingly important conception of computational finance:
the computer becomes an experimental laboratory for markets.
Physics has particle accelerators. Chemistry has wet laboratories. Quantitative finance increasingly has synthetic worlds.
A strategy can be subjected to thousands of hypothetical regimes. A risk system can experience artificial crises. An execution algorithm can meet simulated counterparties. A pricing method can be bombarded with market states that history never supplied.
The essential scientific question then moves from “Did this happen?” to “Could this plausibly happen?”
And that makes validation extraordinarily important.
Alpha is becoming an engineering problem
One of the more surprising statements in the programme appears in a session entitled “From alpha discovery to alpha selection.”
The industrialisation of systematic investing, it argues, has generated such a large universe of possible strategies that the difficult problem may no longer be producing another signal. It may be deciding which signals deserve capital under current conditions.
That is a profound change.
The romantic image of quantitative trading is the solitary mathematician discovering a beautiful anomaly invisible to everybody else.
The modern reality increasingly resembles engineering:
- generate hypotheses;
- acquire and clean data;
- backtest them without fooling yourself;
- estimate capacity and transaction costs;
- combine signals;
- allocate capital;
- execute;
- monitor decay;
- determine when the model has stopped working.
The programme’s crypto session makes the epistemological problem particularly stark. With a single asset such as Bitcoin and only a relatively short useful history, there is little cross-sectional information to average away noise. Consequently the central question becomes whether an observed Sharpe ratio represents genuine structure or overfitting; the proposed tests include parameter sensitivity, comparisons across coins and Monte Carlo experiments.
Quantitative finance is therefore not simply about finding patterns.
It is about proving to yourself that you have not invented them.
Risk has escaped the risk department
The programme also demonstrates how much the definition of financial risk has expanded.
There is conventional market risk, credit risk and counterparty risk. But there is also liquidity risk in decentralised finance, climate and catastrophe risk, cyber risk, supply-chain risk, collateral contagion, model risk, regulatory capital, private-market valuation uncertainty and the feedback effects created when everybody hedges at once.
A particularly interesting DeFi session uses an agent-based model of an Aave-like lending protocol to examine how liquidations, liquidity depletion and user incentives can generate instability and contagion.
This is recognisably quantitative finance, yet there may be no option-pricing formula anywhere in sight.
The object of study is a complex adaptive system.
That same intellectual machinery can apply to a bank, an exchange, a collateral network, an electricity market or a decentralised protocol. Once financial institutions are regarded as interacting computational agents subject to constraints and incentives, the boundary between quantitative finance, economics, computer science and complex-systems research becomes porous.
Even formal mathematics is returning through the computer
At first sight, AI looks like the triumph of approximation.
Yet another strand of the programme points in almost the opposite direction: toward greater mathematical formalisation.
There is a talk on Lean for stochastic calculus. There are arbitrage constraints imposed explicitly during calibration. There are convex formulations chosen partly because they guarantee unique optima. Agent systems are described in terms of auditability and verifiable execution.
This suggests an interesting convergence.
As financial software becomes more complicated and AI makes some components less deterministic, there is increasing value in making the surrounding mathematics and software more formally constrained.
The future quant system may therefore contain an unusual mixture:
- neural networks where approximation is useful;
- Monte Carlo where randomness is unavoidable;
- convex optimisation where guarantees are available;
- symbolic mathematics where exactness matters;
- formal verification where mistakes are intolerable;
- human judgement where no satisfactory formalisation exists.
That is not the victory of one methodology.
It is methodological pluralism.
What about quantum computing?
QuantMinds devotes an entire stage on its final day to quantum computing: quantum annealing for portfolio optimisation, variational quantum eigensolvers for option pricing, hybrid quantum-classical credit models, quantum machine learning for weather and risk, and error mitigation for near-term devices.
The juxtaposition is useful precisely because quantum computing has not displaced conventional computing.
The same day’s programme contains Monte Carlo, neural networks, optimisation, volatility calibration and natural-language processing.
That is probably the correct way to understand quantum finance in 2026: not as a replacement awaiting a ceremonial takeover, but as another computational paradigm being tested against problems whose classical solutions are already highly developed.
The burden is consequently severe. A quantum algorithm must compete not with the computer science of 1990, but with GPUs, specialised numerical algorithms, gigantic clusters and decades of quantitative optimisation.
So what, finally, is quantitative finance?
If the QuantMinds programme is treated as a sample of what practitioners currently regard as important, an answer emerges.
Quantitative finance in 2026 is the discipline of constructing explicit representations of uncertain financial systems and using computation to make decisions within them.
Its ingredients now span:
mathematics + statistics + computer science + machine learning + market structure + risk.
But even this formula misses something.
The most revealing word in the programme may be production.
AI-driven alpha is discussed “from research to production”. Machine-learning models must move from experiments into live systems. Synthetic scenarios must be validated. Agents require harnesses. Strategies must survive transaction costs. Pricing models must generate risk numbers quickly enough for actual desks. Mathematical models must coexist with regulation, infrastructure and imperfect data.
The defining problem of contemporary quantitative finance is therefore no longer merely:
Can we model this?
It is increasingly:
Can we model it, compute it, test it, control it, deploy it and continue trusting it when the world changes?
That is why a conference ostensibly about finance can now contain stochastic calculus, deep neural networks, philosophy of language, Monte Carlo simulation, cloud architecture, quantum algorithms, optimal transport, graph databases, synthetic data and formal verification without actually losing its subject.
They are all different responses to the same problem.
Markets are enormous machines for making decisions about an unknowable future.
Quantitative and computational finance is the science of building smaller machines with which to understand—and act inside—the larger one.