Financial statement analysis in the ERA of big data: A comparative study of machine learning techniques and traditional methodologies

In the modern world of big data, traditional financial statement analysis procedures are being challenged and supplemented with the use of machine learning (ML) techniques. This research aims to provide a comparative analysis of ML techniques and traditional financial statement analysis methodologies in predicting stock market reactions to earnings announcements. The traditional approach is based on financial measurements, particularly profitability ratios (PR). These ratios assist in forecasting stock movements based on past performance. In contrast, the ML approach processes vast amounts of financial and market data using advanced algorithms such as Dynamic Regularization Support Vector Machine with Random Forest (DRSVM-RF). The research uses a dataset of historical financial statements and market data, big data refers to the vast, intricate datasets that require innovative tools for storage, processing, and analysis together with data normalization using min-max normalization and feature extraction using Principal Component Analysis (PCA), to find important parameters like free cash flow and corporate characteristics.  The findings show that the DRSVM-RF model produces the most accurate forecasts and the most abnormal returns. This research indicates that ML techniques outperform traditional methods in predicting accuracy (95.62%), providing a more dynamic and scalable approach to financial analysis, although traditional methods are still beneficial for stable predictions and risk mitigation.