A practitioner-led course on building, testing and defending financial models — from data discipline and statistics through simulation, optimisation and independent validation.
A model is only useful when someone else can reproduce it, challenge it and understand where it stops being reliable. Credit Foncier International teaches quantitative work the way it is reviewed internally: assumptions stated, data documented, uncertainty quantified.
Sessions are delivered by members of our analytics and risk teams. Techniques are built from first principles rather than invoked from a library, and every result is examined for the conditions under which it would fail.
The course is educational in nature. It does not constitute investment advice, does not recommend any instrument, and is not an offer of any financial product or service.
Enquire about a placeEight modules across three levels, from data discipline through to model validation and backtesting.
Before any model is built: sourcing, cleaning and versioning data, documenting assumptions, and establishing the audit trail that makes a result reproducible by someone else.
The statistical foundation that market work actually requires: distributions, estimation, inference, and why financial returns systematically violate the assumptions of introductory statistics.
Linear and multiple regression applied to returns, factor construction, and the diagnostic work required before a coefficient may be interpreted as anything meaningful.
Stationarity, autocorrelation and volatility clustering, with ARIMA and GARCH-family models built step by step and tested against out-of-sample data.
Monte Carlo methods for valuation and risk: generating correlated paths, pricing path-dependent exposures and building scenario sets that a committee will accept.
Mean-variance construction and its practical repair: estimation error, shrinkage, constraints and the risk-based approaches used when expected returns cannot be trusted.
The modelling closest to our own lending practice: probability of default, loss given default, waterfall mechanics and the sensitivity of a facility to recovery assumptions.
How a model earns the right to be used: independent validation, backtesting protocol, overfitting controls, and a sober assessment of where machine learning helps and where it does not.
Build and document a model another analyst can reproduce
Select an appropriate technique and justify why alternatives were rejected
Diagnose a regression or time-series specification before trusting it
Run a defensible simulation and communicate its uncertainty
Validate and backtest a model against overfitting
Every technique is built once in a spreadsheet so the mechanics are visible before any library is used.
Implementation is taught in Python with standard scientific libraries. Prior programming experience is helpful but not assumed.
Published market and accounting data are used throughout, with all the imperfections that implies.
Each module carries a modelling task, read and annotated individually rather than scored anonymously.
Groups are kept deliberately small so that each participant's work receives direct feedback.
The course concludes with an individual model, documented and defended in review, drawn from the participant's own field.
If your question is not addressed here, write to us directly and we will answer it in full.
Write to us with a short note about your background, your current modelling experience and what you intend to build. Places are allocated on a rolling basis throughout the year.