Artificial Neural Networks and Gradient-Boosting Decision Trees in Time Series Forecasting of Earnings per Share in Poland

Precise forecasting of earnings for publicly traded companies holdssignificant importance in achieving investment success. It is highlyimportant in countries where the coverage of these companies byfinancial analysts’ forecasts is relatively small, like Poland. This paperexamines the prediction errors of modern machine learning and deeplearning techniques within univariate time series settings. These meth-ods are applied to earnings per share (EPS) data for companies listedon the Polish stock market during the period spanning from the2008–2009 financial crisis to the 2020 pandemic shock. The seasonalrandom walk (SRW) model achieved the lowest error.