Home / Blog
Blog
Notes from the Riskfuel team
Short takes on AI for derivatives pricing: accuracy, explainability, regulation and what's possible when risk is fast.
How to price thousands of autocallable structures in minutes
Structurers used to price a shortlist of autocallables. With a neural replica, BBVA evaluates thousands of candidate trades in under a minute.
How Riskfuel started: from transistor simulators to derivatives
In 1995, Ryan Ferguson used a neural network to run a transistor simulator in reverse. Twenty years later, the same idea became Riskfuel.
How to speed up IFRS 17 calculations
IFRS 17 runs a full Monte Carlo engine for every contract on each reporting date. Here's how a large Canadian insurer cut reporting from weeks to hours.
Moving XVA to the cloud: what changes when XVA leaves the grid
Banks are thinking about moving XVA to the cloud. Here's why compute efficiency matters even more there and how neural replicas cut the bill.
AAD vs neural networks for derivatives Greeks
AAD speeds up Greeks. Neural replicas speed up the whole valuation. Here's how the two compare and how they work together.
How accurate is Monte Carlo pricing in production?
Production Monte Carlo models carry real numerical error. Here's why that happens and why an AI replica can be both faster and more accurate.
How to run derivatives scenario analysis in real time
Most what-if analysis waits for the overnight batch. Here's how fast pricing lets risk teams run scenarios during the day. Scotiabank did exactly that in March 2020.
Why the best AI pricing combines AI with traditional models
Purely data-driven AI inherits the gaps in its data. Hybrid AI keeps your traditional model as the source of truth and uses AI to make it fast.
Generative AI vs machine learning pricing models
ChatGPT and AI pricing models are both called "AI," but they work very differently. Here's why that matters for banks, risk and governance.
OSFI E-23: what it means for AI and machine learning models
OSFI's revised model risk guideline takes effect May 1, 2027, and it covers AI and machine learning models. Here's what that means for AI pricing.
How to pass the FRTB P&L attribution test
Desks that fail FRTB's P&L attribution test lose the internal models approach. The root cause is two different models. Here's the fix.
What is a neural replica of a pricing model?
A neural replica is a machine-learned copy of your own pricing model. It gives the same answers in a fraction of the time. Here's how it works.
Saving internal capital models
Banks lose the capital benefit of internal models when front-office and risk valuations disagree. One model that's fast and accurate enough for both solves it.
Model explainability
Neural networks have a reputation as black boxes. Three reasons a neural replica of your own pricing model is explainable by design.
Both faster and more accurate
A regulator's report said AI can't match the accuracy of simulation-based pricing. In practice, it can do better.
