Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Attention-based Dynamic Multilayer Graph Neural Networks for Loan Default Prediction

Sahab Zandi, Kamesh Korangi, María Óskarsdóttir, Christophe Mues, Cristián Bravo

TL;DR

A model for credit risk assessment leveraging a dynamic multilayer network built from a Graph Neural Network and a Recurrent Neural Network, each layer reflecting a different source of network connection, bringing better results and novel insights for the analysis of the importance of connections and timestamps compared to traditional methods.

Abstract

Whereas traditional credit scoring tends to employ only individual borrower- or loan-level predictors, it has been acknowledged for some time that connections between borrowers may result in default risk propagating over a network. In this paper, we present a model for credit risk assessment leveraging a dynamic multilayer network built from a Graph Neural Network and a Recurrent Neural Network, each layer reflecting a different source of network connection. We test our methodology in a behavioural credit scoring context using a dataset provided by U.S. mortgage financier Freddie Mac, in which different types of connections arise from the geographical location of the borrower and their choice of mortgage provider. The proposed model considers both types of connections and the evolution of these connections over time. We enhance the model by using a custom attention mechanism that weights the different time snapshots according to their importance. After testing multiple configurations, a model with GAT, LSTM, and the attention mechanism provides the best results. Empirical results demonstrate that, when it comes to predicting probability of default for the borrowers, our proposed model brings both better results and novel insights for the analysis of the importance of connections and timestamps, compared to traditional methods.

Attention-based Dynamic Multilayer Graph Neural Networks for Loan Default Prediction

TL;DR

A model for credit risk assessment leveraging a dynamic multilayer network built from a Graph Neural Network and a Recurrent Neural Network, each layer reflecting a different source of network connection, bringing better results and novel insights for the analysis of the importance of connections and timestamps compared to traditional methods.

Abstract

Whereas traditional credit scoring tends to employ only individual borrower- or loan-level predictors, it has been acknowledged for some time that connections between borrowers may result in default risk propagating over a network. In this paper, we present a model for credit risk assessment leveraging a dynamic multilayer network built from a Graph Neural Network and a Recurrent Neural Network, each layer reflecting a different source of network connection. We test our methodology in a behavioural credit scoring context using a dataset provided by U.S. mortgage financier Freddie Mac, in which different types of connections arise from the geographical location of the borrower and their choice of mortgage provider. The proposed model considers both types of connections and the evolution of these connections over time. We enhance the model by using a custom attention mechanism that weights the different time snapshots according to their importance. After testing multiple configurations, a model with GAT, LSTM, and the attention mechanism provides the best results. Empirical results demonstrate that, when it comes to predicting probability of default for the borrowers, our proposed model brings both better results and novel insights for the analysis of the importance of connections and timestamps, compared to traditional methods.
Paper Structure (24 sections, 7 equations, 10 figures, 13 tables)

This paper contains 24 sections, 7 equations, 10 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Previous work
  3. Graph neural networks
  4. Credit risk modelling with network data
  5. Methodology
  6. Multilayer networks
  7. Topological embedding
  8. Temporal embedding
  9. GNN-RNN models
  10. GNN-RNN-ATT models
  11. Decoder and loss function
  12. Experimental setup
  13. Dataset
  14. Dynamic networks
  15. Experiments
  16. ...and 9 more sections

Figures (10)

  • Figure 1: A multilayer network (left) and its supra adjacency matrix (right).
  • Figure 2: The cell structures of RNN models.
  • Figure 3: GNN-LSTM (left) and GNN-GRU (right) dynamic models.
  • Figure 4: GNN-LSTM-ATT (left) and GNN-GRU-ATT (right) dynamic models. By adding an attention layer to the model, we are able to re-weight the impact of different snapshots.
  • Figure 5: Architecture of the decoder.
  • ...and 5 more figures