Predicting daily changes in the current coupon (CC) rate is central to modelling prepayment cashflows for mortgage-backed securities (MBS).
Traditionally, the LSEG Yield Book platform has relied on two methods: the constant-spread model, which assumes a fixed spread between the CC and the 10-year par swap rate, and the more sophisticated Mortgage Option-Adjusted Term Structure (MOATS) model, which uses Monte Carlo simulations to project rates and prices.
With artificial intelligence (AI) reshaping industries, researchers at LSEG examined whether neural networks (NN) could deliver competitive CC projections while offering flexibility, computational efficiency, and interpretability through techniques such as Shapley Additive Explanations (SHAP). Their study compared the NN approach with both the constant-spread and MOATS models using over 2,000 daily data points from 2016 to 2024.
Performance analysis showed that while NN models provided the best in-sample accuracy, MOATS retained a slight edge for out-of-sample tests.
SHAP analysis further revealed that the 10-year par swap rate remained the dominant predictor for CC changes, but volatility measures contributed valuable insights on specific days, highlighting the advantage of using more complex models for certain scenarios.
The findings point towards the potential integration of NN models into MBS analytics workflows. By balancing performance, interpretability, and efficiency, NNs could complement or even replace traditional approaches, especially for large-scale or real-time pricing needs. Future research will explore whether NN models can maintain performance levels across broader rate and volatility conditions, paving the way for expanded AI adoption in fixed-income analytics.
Find the full breakdown of the research here.
Read the daily FinTech news
Copyright © 2025 FinTech Global
