Automated Data Processing of Bank Statements for Cash Balance Forecasting

The forecasting of cash inflows and outflows across multiple business operations plays an important role in the financial health of medium and large enterprises. Historically, this function was assigned to specialized treasury departments who projected future cash flows within different business units by processing available information on the expected performance of each business unit (e.g. sales, expenditures). We present an alternative forecasting approach which uses historical cash balance data collected from standard bank statements to systematically predict the future cash positions across different bank accounts. Our main contribution is on addressing challenges in data extraction, curation, and pre-processing, from sources such as digital bank statements. In addition, we report on the initial experiments in using both conventional and machine learning approaches to forecast cash balances. We report forecasting results on both univariate and multivariate, equally-spaced cash balances pertaining to a small, representative subset of bank accounts.