Is the inclusion of a broad set of explanatory variables relevant in EPS forecasting? Evidence from Poland

<ns3:p>This study explores the critical role of accurate earnings forecasts for publicly traded firms in achieving investment success, particularly in markets with limited analyst coverage, such as emerging markets like Poland. It evaluates the precision of forecasts generated by a wide array of explanatory variables, including accounting, market, and macroeconomic factors, employing gradient-boosting decision tree machine learning, multilayer perceptron networks, and convolution networks, contrasted witha seasonal random walk model. These models are applied to EPS data from companies listed on the Warsaw Stock Exchange from 2008 to 2019. Multivariate methods are trained using a comprehensive set of 1,598 explanatory variables, encompassing company-specific financial and market metrics along with macroeconomic indicators. The seasonal random walk model demonstrated the lowest error, as measured by the Mean Arctangent Absolute Percentage Error (MAAPE), findings validated throughrigorous statistical examinations. Various robustness checks, employing diverse timeframes and error metrics, reaffirm this outcome. The dominance of a simplistic model may arise from the overfitting tendencies of complex models and the relatively straightforward dynamics observed in Polish listed companies.</ns3:p>