Risk Modeling - Preview Our Technical Paper

#R&D

Ahead of the release of our second technical paper - Methodologies for evaluating credit risk of business data using Machine Learning - our Head of Computational Risk and machine learning expert, Flinn Dolman, shares a preview of what to expect.

Bright is building financial services on the blockchain in order to help small businesses leverage the power of data and improving access to financial services.

Our first step towards fulfilling this goal was designing a cryptographic storage system, the Data Vault. Data Vaults facilitate cryptographically secure data sharing and utilisation. This system is defined and explored in our first technical paper: Data Vaults: Strategies for storing, verifying and sharing private data. For a summary of this paper, read the explainer by our Head of Blockchain, Jaime van Oers.

Our next port of call was to build a system that can fully leverage all benefits that the Data Vault’s architecture sponsors, thus demonstrating the system’s utility. The natural candidate was a Machine Learning based risk model. This risk model will be used to determine credit worthiness of loan applicants to Bright Bank. It will package complex swathes of business data into an easy to interpret credit score that is returned back to prospective borrowers.

In our upcoming technical paper: Methodologies for evaluating credit risk of business data using Machine Learning we discuss why the attributes of Data Vaults give them high affinity to data analysis as well as our scientific approach to data analysis as a whole. More generally, we will explore:

  • Popular approaches to model credit risk in contemporary Machine Learning
  • Metrics for evaluating the effectiveness of a classifier and how the domain we are working within informs the metrics used
  • How Bright addresses the limitations introduced by our data sources and the domain we are working within Bright’s scientific methodology in approaching the ‘credit risk problem’ given the constraints imposed by the domain as specified above
  • Bright’s endeavours for improving both its scientific approach and its process for using data to gauge credit risk of borrowers going into the future

This upcoming paper will lay down the bedrock of the risk model we have developed to inform Bright Bank’s lending decisions. It will do so by illustrating the approach we have taken to analyse the credit risk of Data Vault data. Similarly, it will detail how our methodologies to assess the credit risk of agents on Bright Network will continue to evolve as the project grows and matures.

Be sure to join our Community on Telegram or follow Bright on Twitter to be first to know when this paper will be available. If you have questions, I welcome you to message me on Telegram at @flinnd.

bg