If Multilayer Perceptron Network May Help in Multivariate EPS Forecasting. Evidence from Poland

This investigation delves into the significance of precise earnings forecasts for publicly traded companies in attaining investment success. It highlights the importance of this aspect, particularly in markets with restricted analyst coverage, such as emerging markets like Poland. The study assesses the accuracy of predictions generated by diverse models utilizing distinct sets of explanatory variables, incorporating artificial neural network architectures, in contrast to a seasonal random walk model. These models are employed on earnings per share (EPS) data of companies listed on the Warsaw Stock Exchange spanning from 2008 to 2019. The seasonal random walk model exhibited the lowest error, gauged by the Mean Arctangent Absolute Percentage Error (MAAPE), a finding corroborated through rigorous statistical tests. Numerous robustness checks involving different timeframes and error metrics affirm this conclusion. The ascendancy of a simplistic model may stem from the overfitting tendencies of intricate models and the relatively straightforward dynamics characterizing the Polish listed companies.