Porsenna DSS

BMS Creative product
A portfolio selection system based on Neural Network Advisor

Main Features

PORSENNA

is a modern decision supporting system (DSS) for portfolio selection, analysis and management. It has been developed for banks, broker's offices, trust funds, pension funds, capital investments consultants etc. The main function of the system is to minimize the risk taken during investing on capital markets.

PORSENNA

enables constructing effective portfolios i.e., ones characterized by the highest return-to-risk ratio. Moreover, it allows some fine-tuning in response to the changing market. The interactive process of predicting and modeling the market behavior using the knowledge and experience of analysts and specialists employed at the head department is one of the important features of the system. Together with wide area network technology it results in

the distribution of knowledge

that enables uniform investment policy in all branches as well as gaining a competitive advantage through optimal exploiting of information.

From Neural Model to Effective Portfolio

1. Neural Modeling of the Market

The modeling process, which follows each input of new data (including latest stock exchange quotations) into the system, consists of predicting the return rates of securities with the use of the knowledge acquired by the neural network in the learning process. Acquired predictions of profit together with information on risk and correlation between securities (both indices are calculated parallely) constitute the basis for further data processing and later selection of portfolios.

2. System - Analyst Interaction

At the second stage of processing, there is a possibility of correcting the obtained model by the experts and analysts employed at the head department. This proves to be especially important in cases when e.g., unpredictable market or political events influence stock exchange quotations. Thanks to the Correction Module as well as the processing of selection orders in the central unit the specialists' knowledge is distributed to all branches, which can take advantage of the most up-to-date information.

3. System - Advisor Interaction

Defining the preliminary parameters limiting the selection of a portfolio is a task of the investor or his advisor. Accepted risk level, being the basis for minimal portfolio safety is the crucial parameter. Other requirements are as follows:

• the number of portfolio components,
• kinds of financial instruments (shares, bonds and so on),
• limits concerning each component share in the portfolio,

The present composition of an already existing portfolio becomes an additional parameter in the case of its reconstruction.

4. Presenting the Outcomes

The presentation of portfolio selection or reconstruction results, obtained in course of an interactive process, consists of the following information:

• rate of return,
• per cent shares of components,
• estimated calculations accuracy.

In addition, graphical tools facilitating the comparison of various portfolios are used to visualize presented quantities, thus supporting the decision taking process.

Minimizing portfolio risk

The parameters defined by the investor or his advisor constitute the input information for the portfolio selection module (Portfolio Selector). Adequate securities are selected according to previously set limitations, whereas their per cent shares are chosen in a way securing the highest return rate of a portfolio at a given risk level. The system PORSENNA is based on H.M. Markowitz portfolio selection theory with its later modifications. According to this theory, one can minimize the risk of created portfolios by carefully selecting securities negatively correlated with each other. Portfolios constructed by the system have the most favorable ratio of return rate to risk rate as they are situated close to the so-called effective limit shown in the picture.

Oracle Database and Neural Network Technology

PORSENNA is a client-server system. The grid computing module employs the technology of artificial neural networks supported by modern RDO supervised learning method, and Oracle Database Server ensures high efficiency and reliability of the whole solution.