Predictions of Financial Indicators
Problem
Interest income comes from loans, bonds and credit is the primary way that most commercial banks generate profit. Thus arises the need to assess the ability of the entity receiving funding to repay it, and to assess the possibility of the entity’s default. The most common approach for this assessment is a rule-based one, when the loan auditors on the basis of certain predefined rules, like for example Income above X, decide whether the entity should be given funding. In addition, in some cases, linear models are also used for similar purposes.
Solution
The solutions we offer include a deeper mathematical and scientific approach – building tailor-made ML models. Based on the type of industry, our models can assess the profitability, efficiency, and overall performance of a certain company that requests funding based on the balance sheet and/or income statement financial data. If we take the retail industry as an example, we can make accurate predictions for the company’s inventory drop, inventory turnover, sales forecasting, profitability predictions for the future, and more. The models are built based on the boosted trees approach, which provides a predictive accuracy far more superior than the classical approaches, and helps reduce variance and bias in a machine learning ensemble.
Results
Ensemble models such as LGBM or XGBoost, based on boosted trees, can assess creditworthiness better than the traditional rule-based approach and are more accurate than the linear models. The model results will be especially useful for the borderline scenarios, where the creditworthiness of a company cannot be clearly seen.
In addition, the model implementations help identify which features have the highest impact on the output and provide us with a better business understanding as well as mathematical justification of the process output.
