Machine learning critical for SME credit scoring in trade finance

Machine Learning Credit Analytics for Trade Finance proposes a radical new approach to credit scoring that could particularly benefit SMEs in trade finance.

The paper states that traditional models – such as the Altman Z-score – use a “linear discriminant” analysis, which is based on several accounting indicators. While widely used, such scoring presents a number of issues for SMEs – including focusing on a small number of accounting entries while ignoring valuable non-accounting information. Such hard requirements make credit scoring impossible for companies that miss even one entry. Being based on accounting data filed on an annual basis, traditional scoring also lacks timely information.

The white paper argues that a good predictive credit model for trade finance lending should: accommodate varying data availability across companies to increase the depth of datasets; leverage a broad set of available and emerging data sources, including geographical data; and use trade network data, including common clients, suppliers, or bank relationships, to spot irregularities and predict credit risk.

It’s this approach that will allow for a broader understanding of SMEs’ credit risk, leading to fewer loan rejections and improved credit decisions.

The combination of machine learning techniques with deep and broad data coverage generates a neural network model that can outperform the traditional Altman Z-score and similar models even on pure registration data. And this without using any accounting inputs – hence it’s potentially revolutionary impact on SMEs seeking trade finance.

Tradeteq’s trade asset distribution platform generates credit scoring in just such a way, with the aim of expanding the universe of trade finance investors by encouraging an “originate to distribute” model by trade finance banks. The company – officially launched in March 2018 – is now looking for partnerships and collaborations to work on transaction-level trade finance datasets.