Machine learning–based prediction of long-term farm economic performance: financial, risk, and operational indicators

This study aimed to assess the possibilities of predicting the future development of farms and to identify indicators capable of predicting the success of a farm and its potential risks. This study uses data from Czech farms from 1997 to 2023, which includes financial statements and other production and economic data. Farms are divided into six categories based on a combination of the average four-year return on assets (ROA) and profitability stability. The forecast is carried out using the Random Forest (RF) methodology. The study confirmed the predictability of the future development of farms with an accuracy of 88 %. An essential result of the study is the necessity of using volatility indicators. Operational indicators significantly improve the reliability of the prediction. As regards year-on-year fluctuations in financial ratios, the use of their multi-year average is more appropriate. A recursive representative classification tree was developed and the number of variables needed to classify a company was reduced to five. For the correct classification of a company, the value and volatility of ROA, return on sales, and diversification of production focus are sufficient.

Keywords:
Agricultural enterprises; Financial Health; Profitability prediction; Random Forest