Forex prediction for an investment bank
An investment banking client needed fast, accurate foreign-exchange forecasts to balance risk and return across client portfolios. We replaced their spreadsheet-based process with an AI-driven, web-based system.
The challenge
Their forecasting ran on Excel, which was slow and hard to scale. They needed quicker, more accurate predictions, with their data and proprietary algorithms kept secure.
What we set out to do
- Move from a spreadsheet to an integrated web application
- Replace manual back- and forward-testing with machine-learning predictions
- Protect client data and the proprietary algorithms
- Process large computations quickly
- Improve prediction accuracy
Our approach
Data architecture
We built a BigQuery back end for secure data caching, a secure front end, and a machine-learning pipeline in SQL, Python and R with real-time data collection.
Implementation
We optimised the models, validated them against fourteen years of historical data, and ran beta testing with iterative design.
Results and impact
The client reached insight faster with far less manual data entry, recovered a proper audit trail, and saw a clear improvement in prediction accuracy. It also opened the door to new customers and to other forecasting areas.
“A major improvement. We can now service more customers and report higher accuracy to our investors.”
Timeline
- Initial development: twelve months
- Validation and beta testing: six months
- Ongoing optimisation thereafter
Looking forward
We are extending the approach into sports forecasting and heavy-metals models.
