Investment-grade bonds that cross into sub-investment territory, known as fallen angels, trigger forced selling, portfolio restructuring, and significant loss of capital for institutional holders. Crédit Agricole CIB engaged Multiverse Computing to build a model capable of identifying these downgrade events before they occurred - using quantum and quantum-inspired methods to outperform the classical ensemble strategies already in production at the bank.
The Challenge
Predicting fallen angels is a highly imbalanced classification problem: downgrade events are rare, the feature space is wide, and the cost of a false negative is asymmetric. Crédit Agricole CIB's best in-house strategy relied on an ensemble of 1,200 classical decision trees, which was both computationally expensive and difficult to retrain as market conditions shifted. The bank needed a more compact, accurate model that could meet a demanding precision target at a fixed recall level.