Many studies have predicted the future performance of companies for the purpose of making investment decisions. Most of these are based on the qualitative judgments of experts in related industries, who consider various financial and firm performance information. With recent developments in data processing technology, studies have started to use machine learning techniques to predict corporate performance. For example, deep neural network-based prediction models are again attracting attention, and are now widely used in constructing prediction and classification models. In this study, we propose a deep neural network-based corporate performance prediction model that uses a company's financial and patent indicators as predictors. The proposed model includes an unsupervised learning phase and a fine-tuning phase. The learning phase uses a restricted Boltzmann machine. The fine-tuning phase uses a backpropagation algorithm and a relatively up-to-date training data set that reflects the latest trends in the relationship between predictors and corporate performance.

Abstract:Many studies have tried to predict corporate performance and stock prices to enhance investment profitability using qualitative approaches such as the Delphi method. However, developments in data processing technology and machine-learning algorithms have resulted in efforts to develop quantitative prediction models in various managerial subject areas. We propose a quantitative corporate performance prediction model that applies the support vector regression (SVR) algorithm to solve the problem of the overfitting of training data and can be applied to regression problems. The proposed model optimizes the SVR training parameters based on the training data, using the genetic algorithm to achieve sustainable predictability in changeable markets and managerial environments. Technology-intensive companies represent an increasing share of the total economy. The performance and stock prices of these companies are affected by their financial standing and their technological capabilities. Therefore, we apply both financial indicators and technical indicators to establish the proposed prediction model. Here, we use time series data, including financial, patent, and corporate performance information of 44 electronic and IT companies. Then, we predict the performance of these companies as an empirical verification of the prediction performance of the proposed model.