A financial early warning system for financial intermediary institutions by Neural networks

Abstract

The 2008 economic crises revealed that the existing financial system requires better monitoring and more effective regulations of the financial institutions. Straightforward implementation of tighter regulations will increase the costs of the financial system which will eventually hurt economic development. In order to minimize the effects of tighter regulations on the costs, regulators shall also consider taking advantage of new methods which are more complicated than existing ones. This dissertation proposes a financial early warning system for broker dealers in Turkey. Discriminant Analysis and Neural Networks are used comparatively and cooperatively to develop the model tailored for broker dealers. An extensive database is formed by Capital Adequacy Reports that were collected by Capital Markets Board for the period between 1999 and 2009. Access to this database contributed to this study in many ways through its tailored structure truly reflecting the financial standings of this industry. Popular independent variables in the literature are used and new ones are also proposed in order to take advantage of the details in the extensive database. Discriminant Analysis is used to elect the important independent variables that formed the backbone of the model, although most of the important a priori assumptions were violated. Neural Networks picked up from where Discriminant Analysis left and final model provided approximately 75% classification accuracy. Such a figure may seem low compared to similar studies. However the model predicts the deficiency in the capital adequacy, a pre-default event, which is obviously more difficult to predict than default itself.