ML-Based Predictive Modelling of Stock Market Returns

Many recent papers conclude on short-term profitable patterns based on trading strategies relying on the belief that best (worst) performing securities over the past short-term period tend to continue to perform well (poorly) over the subsequent period of up to 12 months. Along with exploiting past price patterns, ever since the early 1980s the researchers' attention is drawn on the usefulness of company financial statements in predicting stock market returns. We argue on the necessity to develop a framework that would enable investors to support their everyday decisions by taking into account information retrieved from financial statements in an automated manner We propose an ML-based AI system that inputs historic price information as well as information on a number of balance sheet items, profit and loss items, and cash flow items for the company of interest and predicts the probability of its stock price going down in the next period. The prediction relies on a dynamic automated selection of significant accounting and historic features. Tracking the dynamics in the structure of significant features could shed additional light on company specific issues that might be taken into account by investors when making their decisions.