AN INVESTMENT PREFERENCE UNDER INCOMPLETE DATA I. Popchev and I. Radeva Institute of Information Technologies Akad Georgy Bonchev str., bl. 2 1113 Sofia, Bulgaria Tel.: (+359 2) 716851, Fax: (+359 2) 9434589 Email: iradeva@iit.bas.bg Abstract: The presented approach uses contradictory results of different multi- discriminant bankruptcy prediction models, applied to a sample of Bulgarian public companies. It allows the application of an expert classification of the sample investigated. Conditions are provided for discriminant analysis appliance, through a grouping variable based on an expert decision and named Investment Preference. The approach presented enables an analysis of the Investment Preference under conditions of incomplete data. Copyright © 2004 IFAC Keywords: Economics, Incomplete data, Evaluation, Decision making, Discriminant analysis, Iterative method 1. INTRODUCTION 1 The techniques based on fundamental and technical analysis are widely applied in investment decision making on capital markets. The technical analysis on Bulgarian market, although already applied turned out to be not quite effective as it is not liquid enough, the market capitalization is low and free floats of listed companies are few. The fundamental analysis is accepted as a preferred technique, but it is costly and hardly affordable for individual investors. Alternatively, the financial ratios analysis mɚy be used as the input data to perform it are accessible and many software programs offer its calculation. The bankruptcy prediction modeling is also an acceptable tool for investment decision making. This field is widely discussed in economics literature. Three basic techniques are used for developing bankruptcy prediction models (BPMs): Discriminant analysis (DA), artificial neural networks (ANN), and 1 This work is supported by the National Science Fund of the Bulgarian Ministry of Education and Science under grand No. I1305/2003 fuzzy sets (FS). Regarded pioneers in developing models based on DA are Beaver, Altman ɢ Ohlson. Bankruptcy prediction based on ANNs begins at 90’s of the 20 -th century and marks a rapid development (Atiya, 2001). Ⱥs a whole, the ANN models’ classification accuracy is comparable with that of the Multivariate DA models. The latest techniques applied in bankruptcy prediction modeling are FS. Each of the techniques afore mentioned has its advantages and shortcomings. Anyway, DA remains the leader with respect to range of applicability (Shirata, 1998). Therefore, in this paper, DA was the preferred techniques to achieve the objectives of this study. The objectives are: 1. To verify an applicability of some known BPMs over a sample of Bulgarian companies. 2. To develop an approach for expert classification of public companies under conditions of incomplete data. 3. To apply DA for verification of the expert classification in point 2 and for actualizing of this classification. IFAC DECOM-TT 2004 Automatic Systems for Building the Infrastructure in Developing Countries October 3 - 5, 2004 Bansko, Bulgaria