Applying Machine Learning Algorithms to Predict Liquidity Risks

Liquidity risk is a significant financial threat, and its mismanagement can lead to substantial financial losses.This study examines the application of machine learning techniques like KNN, SVM, decision tree, RF and XGBoost for predicting liquidity risk in Indian banks.Financial data from 2013-2022 for 31 commercial banks is analyzed.The models use financial ratios as predictors and liquidity risk is proxied by liquid assets to total assets and loan to deposit ratio.Despite limitations in generalizability due to the small sample, the results exhibit the potential of algorithms like KNN and XGBoost for accurate liquidity risk forecasting.The study's findings revealed that the model that used liquid asset to total assets to proxy liquidity risk gave the best results with the KNN, and gave MAE and MSE scores of 0.129 and 0.027, respectively.When loan to deposit was used to proxy liquidity risk, DT was the best-performing algorithm, with an MAE and RMSE score of 0.191 and 0.231, respectively.It was also found that MLP did not perform well when compared to other selected models.The practical implications include developing liquidity early warning systems for Indian banks using these techniques.