Scientists from Tomsk Polytechnic University (TPU), as part of a research group, have developed a new method for predicting the profits of high-tech companies. The accuracy of forecasts has increased by 30% due to the use of machine learning algorithms.
The combined method for predicting profitability is based on machine learning, regression with random effects, and clustering. Associate Professor Nikita Martyushev said that the method breaks down profit into three components: trend (normal profit growth), seasonality (profit growth in a specific period), and residual component (profit that cannot be predicted).
Before the new method was applied, the average forecast error was 5%, but after implementation, it decreased to 3.5%. This was due to a detailed analysis of seasonal fluctuations and the identification of hidden trends. The use of machine learning algorithms reduced the root mean square error of profit forecasts by 25% compared to traditional methods.
The study involved scientists from the School of Information Technology and Robotics Engineering and the TPU Business School, Moscow Polytechnic University, and Irkutsk National Research Technical University. They analyzed the financial and operational performance of 1,811 companies for the period from 2013 to 2018.
Co-author of the study, Lyubov Spitsyna, announced plans to test the method on enterprises in other sectors of the Russian economy. This will allow us to assess the applicability of the methodology in various industries and compare management approaches.
The research results are published in the scientific journal Mathematics.
Read more materials on the topic:
A system for predicting stock price growth has been developed in Moscow
A new approach to predicting the future has been developed in Russia
Novosibirsk scientists have developed an algorithm for predicting exchange rates and weather
