Networkbased credit risk models in P2P lending markets

A hands-on approach to establishing trust between investors and P2P markets is to use accurate credit risk models. The objective of the research is to design a state-of-the art and interpretable credit risk models for P2P lending markets.



The proposed research project focuses on the development of new credit risk models for P2P credit markets enhanced by features from networks. We will test and compare the usefulness of new models with several real data sets from P2P credit markets worldwide. For a credit risk model to be useful, we also need to consider the interpretability of our models. Therefore, our research project has both methodological and empirical features with practical implications. The main objective of the research project is to deepen our understanding of credit risk modelling in P2P credit markets by designing and empirically testing new network-based credit risk models. The research project will lead to the following contribution: Methodologically, we aim to contribute to several aspects compared to existing research (Ahelegbey et al., 2019a,b; Giudici et al., 2019; 2020). Using network-based learning for machine learning models, we aim to exploit composite properties of loans in portfolios to uncover additional information for credit risk assessment. Supervised learning algorithms are generally more powerful than unsupervised learning algorithms (Liu et al., 2020). Therefore, our approach to network construction will use class information, resulting in supervised networks and should lead to an advance in credit risk projection.

Course of action

Here are some ideas that show how networks can help us develop better models for assessing credit risk in peer-to-peer credit markets. Previous models have used a technique called minimum spanning tree (MST), which greatly reduces many relationships between points in the network. This could lead to losing important information. Also, the method is designed to connect all points in the network, but this does not necessarily make sense in terms of credit risk. We want to try other types of networks where we can keep a certain percentage of the closest connections. In simple words, we want to find new ways to use networks to better understand how big the risk is that someone will not repay their loan. In doing so, we want to make sure that our models really provide useful information and not just complicated mathematical connections.


This research aims to revolutionize P2P lending markets by developing advanced, interpretable credit risk models. By utilizing innovative network-based learning and supervised networks, we intend to delve into the composite properties of loans to unlock additional information crucial for meticulous credit risk assessment. Our endeavor to exploit various network types is designed to preserve pivotal information and to refine credit risk projections significantly. The anticipated outcome is the establishment of advanced, user-friendly models that represent valuable, actionable insights, enhancing understanding of the risks in loan repayments and making P2P markets more secure and reliable for investors, ultimately fostering a higher degree of trust between investors and P2P markets.

Looking ahead

By using network-based approaches, we aim to gain deeper and more precise insights into risk factors. Previous models may have missed important relationships and information; our approach aims to fill these gaps. In particular, we aim to develop models that are not only mathematically complex but also interpretable and useful in the real world. A particular focus is on designing the models to accommodate a variety of data sources while being flexible enough to adapt to different market conditions. By using multiple datasets from different P2P lending markets around the world, we aim to confirm the robustness and applicability of our methods.