Imagine a world in which the value of receivables could be predicted so accurately in almost every instance that finance companies and factors would know precisely on which of a company’s invoices it was safe to advance, and the risk of collection had shrunk to a negligible amount. SMEs trading across the world could access more working capital unproductively “stuck” in unpaid invoices more readily and at a lower cost than ever before.
Such a world is closer than you think, says Michael Boguslavsky, Tradeteq’s head of AI, thanks to a fresh look at the data required for the credit analysis of SMEs, and enlisting the help of AI, as explained in his newly-released whitepaper Machine Learning Credit Analytics for Trade Finance.
Traditional techniques lack flexibility
Despite their requirements often being relatively modest, poor credit scoring is a key reason many SMEs fail to win the trade finance they need. This applies across the board, not just to receivables finance. A large part of the problem lies in the traditional approaches to credit scoring.
Many small companies cannot satisfy the narrow requirements of traditional credit rating techniques such as the Altman Z-score, introduced in 1968. With its improved versions now widely used, the Z-score uses linear discriminant analysis based on a highly-selective number of financial metrics, ignoring potentially valuable accounting and non-accounting company information. A company who can’t supply even one of the required accounting entries is not rated at all, so many perfectly credit-worthy SMEs are simply rejected. Worse still, the accounting data on which the Altman Z-score is based are taken from company accounts filed annually and consequently are often out of date.
Receivables due from SMEs suffer under the same impediment. Finance companies giving them a credit score using traditional techniques reject or heavily discount many small companies’ invoices that deserve a better credit score (and may be advancing on some that should be rejected). While attempts have been made to improve on traditional techniques, Altman Z’s inability to provide accurate credit-scoring for SMEs remains.
New approaches are now being developed, using machine learning and drawing on far richer, more diverse and more up-to-date data sets. These models predict the likelihood of credit events occurring for different types of company with significantly greater accuracy than Altman Z-score-type approaches.
Wider and deeper data
By using a broad set of available and emerging data sources, such new models exploit categories of company information previously uncaptured. For example, mapping a company‘s registration address to socio-economic area classification and census data is just one way non-financial information can be relevant to analysing its credit status. Industry data or a company’s history of renaming and mortgage charges are other data sources to be used, and the list goes on. Also, unlike under the Altman Z-score, there is no hard requirement for a limited number of accounting entries which excludes all companies failing to provide them. Instead, new models accommodate varying data availability across companies, utilising all available data, noting absences and learning from the resulting patterns.
New models also use more up-to-date data than those filed annually – or even less frequently – in a company’s accounts. More frequent and more detailed private data sets e.g. from banks, large customers, electronic invoicing companies and digital marketplaces, can reveal underlying credit exposures linked by common clients, suppliers or bank relationships. The data are therefore richer than those traditionally used, leading to a better-rounded and more nuanced understanding of SME behaviour and credit risk.
Improves credit scoring accuracy with machine learning
Another key factor is AI, using neural networks on data sets to assess inter-relationships between different factors. This dramatically improves the quality and timeliness of credit-event prediction for different types of company. The joint use of better quality data and improved prediction techniques allows the new neural network models to outperform traditional Altman Z-score (and similar) models even on pure registration data, without using any accounting inputs. When using these models to score SME receivables, it is equally important that they have much wider coverage than the Altman Z-score models. Those models excluded smaller companies on the basis that their financial ratios were too unstable for a failure prediction model ‒ new models never reject a company because its assets are too low.
The latest large-scale test of the Altman Z-score in 2014 covered private limited companies in 35 countries, including around 340,000 companies in the UK – just 13% of all registered limited companies. By comparison, Tradeteq’s current model combines inputs from Companies House, the London Gazette and the Office of National Statistics and provides credit scores for all 3.4m UK active limited companies. On this dataset, the Altman Z-score’s key “AUC” metric (used to measure model performance) was between 0.70 and 0.74, whereas Tradeteq’s Neural Network UK limited company credit model compared very favourably with an AUC of 0.92. This means it delivers significantly more accurate predictions of the likelihood of SME credit events.
A feedback loop of growth
Using these new models, receivables finance becomes more accurate, less risky, making it a more readily available and less costly source of working capital for SMEs than ever before. In addition, the acquired invoices become more valuable, attracting additional cashflows into receivables finance and fuelling a virtuous circle of growth in trade and prosperity. Tradeteq is now looking for partnerships and collaborations to work on transaction-level trade finance datasets, leveraging its expertise in deep data analysis and the broad data sourced from partners to produce state-of-the-art credit analysis for the trade finance community.