Multi-agent Architecture for Corporate Operating Performance Assessment

Due to great changes in the global economy, corporate financial distress prediction is playing an increasingly vital role in this highly competitive environment. However, one key factor of financial distress is poor management, and business operating efficiency is a suitable reflection of corporate management. The multi-agent architecture introduced herein, namely the ROER model, consists of three main parts for corporate operation efficiency forecasting: (1) data envelopment analysis with random projection, (2) online sequential-extreme learning model (OS-ELM), and (3) rough set theory (RST). The ROER model is grounded on ensemble learning/multi-agent learning, of which the core elements are preciseness and diversity. We achieve the goal of generating a diverse model through two ensemble strategies: (1) modifying the inherent parameters of RP, and (2) adjusting the activation type, block size, and parameters of OS-ELM. The nature of the multi-agent architecture is opaque, making it complicated for users to comprehend as well as impeding its empirical application. To handle the black-box problem, this study implements RST to extract the decision knowledge from the multi-agent model and to represent knowledge in an ‘if-then’ format, which is easier to understand. In the modern economy, the essential component of any value-creating business activity has changed from physical assets to intangible assets. Thus, we examine intangible assets, with empirical results revealing that the proposed ROER model is a promising alterative for the performance forecasting task.