Quantitative finance consulting
Models that hold up under scrutiny
JoaFin builds quantitative models — statistics, machine learning, risk and valuation — and the backend that puts them to work in the hands of a trading desk. We use AI heavily to move fast; what we deliver rests on mathematics we can derive, document and defend in front of a committee, a regulator or a court.
F_μ and F_ν are the distribution functions. On the line, with convex cost, the optimal map is the monotone rearrangement — which is exactly what the figure computes.
What we bring to the table
Statistics and machine learning
Predictive modeling, uncertainty quantification and validation. With published research on when a learning model converges and when it does not.
Mathematical finance
Derivatives, local volatility, pricing and numerical methods running in production.
Risk and heavy tails
Extreme value methods, daily CVaR limits and statistical arbitrage traded on a desk.
Energy markets
Locational marginal price forecasting and daily trading in the Mexican power market.
PhD in Mathematics
University of Toronto, 2024, under Robert J. McCann. Analysis, probability and optimal transport.
Backend for quant teams
From data acquisition to the model the desk queries every day.
The starting point
AI accelerates the work. Mathematics is what holds it up.
JoaFin exists for a specific reason: a lot of people are using AI badly. They use it to skip past framing the problem rather than to reach a well-framed one sooner. Our bet is the quantitative consulting that always worked — models, risk, valuation, infrastructure — amplified by these tools rather than replaced by them.
A quantitative model is not worth what it cost to write. It is worth what can be asserted about it: the conditions under which it converges, the assumptions it rests on, how sensitive it is to the data, and where it stops being valid.
We use language models every day to implement, review literature, prototype and automate. It is a genuine speed advantage and we take it without apology. But problem formulation, choice of method, error bounds and verification are ours, and they are not delegated.
On this we are not speaking second-hand. One chapter of the doctoral thesis this practice is founded on is precisely about when a machine learning model converges under modification of its data, and when it does not. The proof is published. We know where these tools break because that was the object of study.
The practical consequence: you get in weeks what traditionally takes months, without giving up traceability of a single assumption.
How we workServices
Seven distinct areas, one criterion: the method is chosen for what can be proven about it.
- 01
Statistics and machine learning
Predictive models built to be valid, not merely to fit well in sample. This is the area where apparent performance is most often mistaken for real performance.
- 02
Mathematical finance and derivatives
Pricing, calibration and numerical methods for volatility surfaces and nonlinear structures. This is day-to-day work on the team.
- 03
Risk, heavy tails and statistical arbitrage
Risk measurement where Gaussian assumptions fail — which is precisely where risk matters.
- 04
Power and energy markets
Nodal price, demand and risk models for participants in the Mexican power market and international counterparties.
- 05
Data, backend and productization
Most of the work ends up here: the model has to run on its own and sit within reach of whoever decides.
- 06
Actuarial modeling
Valuation of contingent liabilities, decrement tables and stochastic projections with explicit assumptions.
Selected work
Representative projects. Client names are withheld for confidentiality; the technical detail is open to discussion.
Energy markets · Trading
Daily statistical arbitrage and heavy-tailed risk limits
An energy desk was running with risk limits calibrated on Gaussian assumptions. In power markets the distribution of outcomes is nowhere near Gaussian: the episodes that define the year — severe congestion, scarcity, heat waves — live in the tail, and that is exactly where a badly set limit stops protecting anything.
Derivatives · Analytics platform
Optimal-transport-based local volatility calibration
A derivatives analytics platform needed local volatility calibration fast and stable enough to live inside the desk's decision cycle — not as an overnight batch — and surfaces that were arbitrage-free by construction rather than by later inspection.
Power market · Mexico
Locational marginal pricing models
Participants in the Mexican power market needed to anticipate nodal prices to size positions and evaluate hedges, in a market with abundant but irregular public data where congestion dominates the variation between nodes.
Data · Backend for desks
From raw source to the model the desk queries
Analysis that depended on data scattered across public portals, files and email, and models that lived in one person's notebook. Every query was a favor and every result was irreproducible.
Have a problem that spreadsheets won't settle?
If the problem is quantitative, ill-posed or simply hard, the first conversation usually clarifies more than expected. No cost, no obligation.