Financial analysis using Bayesian networks (Q2722290)

Constructing a Bayesian network for a given variable domain could be done in two ways. One way is to construct a network ``by hand'' using uncertain expert knowledge. Unfortunately, this procedure is obviously tedious and time consuming. The other approach is through automated learning using observed data. In this paper these two approaches are combined: encode expert knowledge in a Bayesian network structure, then use this knowledge to improve speed and accuracy of learning using data. Only one part is chosen (fundamental analysis of enterprises) for the focus of this paper. The object of the fundamental analysis is to determine an enterprise's economic situation by means of data gathered from within the enterprise itself. The paper is concentrated with the analysis of the financial indexes of one economic sector: engineering. This choice is motivated by the relative stability in the Czech Republic, compared to other sectors, and by the fact that there is at the disposal a sufficient number of annual reports. About 4400 reports of engineering enterprises for the year 1993-1997 were collected, from which 10 basic financial indexes were collected which are typically used by bank analysts to judge an enterprise's application for a loan.