Information Technology | Studies, essays, thesises » Dr. Baracskai-Ormos - Managing risk in investment decision by Doctus knowledge based system

Baracskai Z, Ormos M, Velencei J., “Managing Risk in Investment Decision by Doctus KnowledgeBased System” 12th Annual Entrepreneurial Finance and Business Ventures Research Conference New York, USA, 2001.0427-20010428 New York: Syracuse University School of Management, pp 1-14. Managing Risk in Investment Decision by DOCTUS Knowledge-Based System Dr. Zoltán BARACSKAI Associate Professor at the Budapest University of Technology and Economics, Faculty of Social and Economics Sciences, Department of Industrial Management and Business Economics. E-mail: baracskai@imvt.bmehu, Tel: (+36)-1-463-1092, Fax: (+36)-1-463-1606 Address: H-1111, Budapest, Mőegyetem rkp. 9 bld T Mihály ORMOS Ph.D student at the Budapest University of Technology and Economics, Faculty of Social and Economics Sciences, Department of Industrial Management and Business Economics. E-mail: ormos@imvt.bmehu, Tel: (+36)-1-463-1607 Fax: (+36)-1-463-1606 Address: H-1111, Budapest, Mőegyetem rkp. 9 bld T Jolán

VELENCEI Scientific Researcher at the Budapest University of Technology and Economics, Faculty of Social and Economics Sciences, Department of Industrial Management and Business Economics. E-mail:velencei@imvt.bmehu, Tel: (+36)-1-463-1092 Fax: (+36)-1-463-1606 Address: H-1111, Budapest, Mőegyetem rkp. 9 bld T 1 ABSTRACT Decision on an investment is a typical multi criterion decision (MCD). In this paper we try to find logical relations between the criterions. Based on these relations the evaluation of the project is deduced. So this is a typical deductive conclusion. For deduction in Doctus the knowledge is represented in the form of ‘If A then B’ rules. Classifications of cases and output of rules can have not only single value, but also distributed values based on their probability. Investment analyses include numerical and non-numerical attributes, both of theses has to be integrated in to the decision tree. The events occur in the environment is uncertain, so based on

experts’ intuitions a level of risk is determined. Risk means some kind of deviation form the expected value of attribute values. The last section of the paper deals with the application of Doctus knowledge based system for investment decision in uncertain environment. BACKGROUND In deduction the conclusion is unknown. The expert gives the presumptions and the rules therefore this is also called Rule-Based Reasoning (RBR). For deduction in Doctus the knowledge is represented in the form of ‘If A then B’ rules, usually called production rules. “To avoid a redundant storage of all dependencies between A’s and B’s, deduction principles are available. If situative conditions or facts become evident, desired answers can be entailed by these principles from stored and from deduced knowledge.” (Rödder, 2000) In Doctus only one of these principles is used. The expert and the knowledge engineer build a hierarchy of presumptions so the number of rules could be reduced and

therefore they are easier to handle. We call this hierarchy a deductive graph Building a knowledge base The first step is the acquision of attributes. Cases in Doctus KBS are characterized by attributes. Attributes are actually points of view, in connection with which we have 2 expectations in relation to cases. The definition of attributes essentially provides a measurement of the knowledge base. The primary condition of rule based reasoning is that attributes should be subordinated to each other with given rules. The deductive graph describes the dependency relations of attributes. Nodes depend on nodes (factor attributes) connected from below Deduction is, technically, a matter of upward tracking on the graph. The top of the graph is the final conclusion. On the basis of subordination, two types of attribute can be distinguished: attributes depending from other attributes are called dependant attributes. The values of these are determined by rules given by user. Attributes

which do not depend on anything are called input attributes, or independent attributes. The values of these attributes are given by user. The second step is acquision of rules. A rule determines the value of its attribute for a given value combination of factors. So the domain of rules is, therefore, the complete range of possible permutations of a stock of factor values. By way of illustration: this range is a k dimension field, the extent of which is determined be the number of factors (k). Rules are placed in this field so that, in good cases, they will fill it completely, and so every case is covered. Rules, which are valid for just one selection of factor-values, are called elementary rules. The valid range of a rule can be more extensive, in as much as a certain factor value may be of concern to other neighboring values. These rules are called complex rules They can be considered as a rule assembled from elementary rules. 1. Figure: Complex rules 3 2. Figure: Illustration

of the above rules The valid range of complex rules can be illustrated by a k dimensioned parallelepiped. In our example, k=2. In accordance with this, the first two rules are 1-dimensional lines; the others are two dimensional ‘planes’. The extreme case of complex rules is the complete rule, which is valid for all values of its factors; in this way it fills out the whole domain. In the example, the 5th one is such a rule. In addition, an output value may be ascribed to each rule, which becomes the value of the attribute in the course of deduction, in as much as the actual values of factors fall within the valid range of rules. Storage: the rules create a list, allowing cases to be uncovered and, moreover, multiple definition of one or more ranges. In this case the first rule is considered valid Thanks to this, the simple solution of the ‘otherwise rule’ is that all the undefined ranges can be covered with a single complete rule placed at the end of the list. Wholly hidden

rules, and even the partial overlapping of a range can be revealed. The ordering of rules has, therefore, a definite role, since the reordering of rules can alter the output of a rule set. There are two possible way of displaying the rules the first is “list of Rules” It initially shows the rules as the shell stores them. They are arranged line by line, with the first at the top The rightmost column contains the output value for each rule. Left of this can be seen the values or value ranges of factors that define the valid range of rules. This tab is used for editing rules directly. The second way of visualization is the “two dimensional” approach It is possible, at any time, to switch over to another elevation where the k dimensional domain can be seen flattened out into 2 dimensions (2D). Essentially, 2D slices can be seen either 4 one under the other, or side by side. Individual cells thus represent one or another selection of factor-values, in other words one or

another point of the domain, regardless of whether it is covered by a rule. If it is so, the output value of the active rule that concerns the given case is displayed. Complex rules with an extended valid range can, of course, occupy more cells (see figure 9). 3. Figure: Illustration of figure 9 On the other hand, slices can be rearranged easily: the cell for the name of factor-attribute (top-left corner) can be dragged, and its head moved into another position. The widest arrangement that can fit in window can be found automatically. One of the most useful option is checking consistency. The basis of Doctus is that the values of attributes can usually be arranged according to their worth and, from amongst 2 ranges, the output of the better ones cannot become worse, nor vice-versa. Thus it is only needed to enumerate values by ordering, and to give the direction of ordering. Eg The values of the ‘buying price’ attribute have a quantitative worth: the order of “expensive –

acceptable – cheap” is increasing; however, the values of the “colour” attribute cannot be ordered (i.e it is nominal). In this sense, it has many roles: checking consistency of given rules, looking for contradictions, as well as the displaying and acceptance of advice for uncovered ranges. Consistency is not obligatory, of course. The control can be switched off for any attribute in as much as the order of values can be set as «nominal». If it is only needed to inactivate it 5 for certain dimensions, then the corresponding factor has to be modified; to disable checking of whole rule list, “nominal” is selected for the corresponding attribute. The risk in Doctus is managed by its distribute function. Classifications of cases and output of rules can have not only single value, but also distributed values based on their probability. INVESTMENT ANALYSES BY DOCTUS KBS Attributes In this case the decision on a investment implementation was made by the model below. It

can be seen that the implementation depends on the economic valuation of the project, the functional value of the project and as the project will be somehow environment polluting we had to take into consideration the environment protection as well. Economic Valuation IMPLEMENTATION Functional Value Environment Protection 4. Figure: The implementation of an investment depends on The economic valuation of the project was carried out by four considerations, these are: • the net present value (NPV) - which can be much higher than zero (the project from economic view can be accepted,) equal to zero or just slightly differ from zero (the project can be take into consideration) or negative (in this case the project is not economical so it has to be rejected from this point of view) • the estimation of the variables in the NPV – while the net present value depends on the estimations of the future cash flows indicated by the project and 6 on the estimation of the opportunity

cost, the quality of these estimations has to be taken into consideration, • Investment cost – if the investment cost is not too high then the risk will not be too high as well, so the model allows us to bear with it – and • The limit of capital – if the capital is limited only the best project can be realised. Net Present Value reject (<0) considerable acceptable (>0) Cash Flow Estimations not accurate acceptable accurate NPV Estimation not accurate acceptable accurate Economic Valuation Not econimical considerable acceptable Estimation of the Opportunity Cost difficult estimatable Investment Cost high medium low Limit of Capital exist Not exist 5. Figure: The economic valuation of an investment depends on The functional value was associated with the result of the investment and examined by the forthcoming viewpoints: • The product in this case was judged upon its quality, synergy with other product in the portfolio, and upon the product references •

Image of the product and the effect of the new product on the company image. Sinergy Product Product References Image Product Quality Functional Value 6. Figure:The functional value of an investment depends on 7 The environment protection was judged by two attributes: • If the project is necessary but environment polluting usually the government or other international funds gives additional financial assistance, the presence and the value of this was examined under this point • In this type of case environment protection standards exist. Here the correspondence to the standards can be defined. Additional financial assistance Environment Protection Standards 7. Figure: The level of the environment protection of an investment depends on So the whole model (deductive graph) shows the next picture: Net Present Value Cash Flow Estimations NPV Estimation Economic Valuation Estimation of the Opportunity Cost Investment Cost IMPLEMENTATION Limit of Capital Sinergy

Product Product References Image Product Quality Functional Value Additional financial assistance Environment Protection Standards 8. Figure: Deductive graph on the implementation of an investment Attribute values The next step after determining the above attributes of the implementation, the attribute values. The values of attributes are grades in the domains given by the experts In this case the values of the above mentioned attributes are shown in the next figure: 8 9. Figure: Attributes and their values Rules Giving the connection by “if then” rules between the above mentioned parameters, a small knowledge base presented by the experts. Only the rules of the NPV estimation will be presented this time. On the upper half of the figure () the rules can be seen in a two dimensional way e.g IF difficult to estimate of the opportunity cost AND the Cash flow estimation is accurate THAN the NPV estimation will be acceptable by the opinion of the expert. On the lower part

of the figure the same rules can be seen in a one dimensional way e.g in the second row compact rule can be found It says IF the opportunity cost estimation is difficult and the Cash flow estimation is acceptable OR worse THAN the NPV estimation is not accurate. 9 10. Figure: Rules of NPV estimation Cases In this presentation four different projects will be described. The input parameters can be found in the next table: Image Limit of Capital Synergy Product Quality good Not exist positive high Case 2. average Not exist positive high Case 3. Case 4. negative high positive average Case 1. good good Not exist Not exist Additional Estimation of Invest. Cash Flow Net Present Product Standards financial the Opportunity Cost Estimations References Value Cost assistance acceptable estimatable high accurate known adequate low (>0) acceptable difficult low accurate hardly known adequate low (>0) considerable estimatable low accurate no any adequate high reject

(<0) estimatable high accurate well known adequate no Case 1. In this situation the implementation of the project is recommended, because the economic valuation of the project is acceptable – however it requires high investment cost but the NPV is positive, its estimation is accurate and there is no limit of capital – the functional value is high – the product is excellent and the image is good – and the environment protection means a small problem –because there is no additional financial assistance. 10 Case 2. Case 2 is OK from the view of implementation. In this case the Knowledge based system suggests that the parameters are good enough to implement the project, however in some extent it shows disadvantages comparing to Case 1. Eg the functional value is just considerable because the image is average and the product is acceptable (the product is only hardly known). Case 3 The project is implemented just in the case when there is no any better solution and with

care because in many attributes it shows weaknesses. However in sense of functional value and environment protection this project is better than Case 2, the economic valuation is just considerable because the net present value is just considerable i.e close to zero Case 4 Case 4 is not recommended using the rules given by the experts. Although the functional value is the highest from the presented cases the other attributes are much worse than in other ones. The figure below shows the deductive graph with the Cases and their attribute values, so one can follow the “decision flow” procedure through the values and the rules. 11. Figure: Deductive graph with the cases 11 Managing Risk By Doctus Risk in this situation means some kind of deviation form the expected value of attribute values. This shows the main uncertainty in investment decisions (This was tried to be eliminated when the NPV estimation as an attribute have been involved in the model above, however this cannot be

enough to handle uncertainties). Eg in case 4 the future product quality is now thought to be high, however there can be e.g 15% possibility of that will be only average and 5% of low quality. This can be handeled by the distribution function of Doctus. The affect of this to the previously introduced cases will be shown on the next figure 12. Figure: Distribution function This shows that the expert defined the probability of different values of attributes within the four cases. It is obvious that the output of the model has to change according to the possibilities. 12 Net Present Value 1 - acceptable (>0) 2 - acceptable (>0) 3 - considerable 4 - reject (<0) Cash Flow Estimations 1 - accurate 2 - accurate 3 - accurate 4 - accurate NPV Estimation Economic Valuation 1 - acceptable 2 - acceptable . 3 - considerable 4 - Not econimical 1 - accurate 2 - acceptable 3 - accurate 4 - accurate ost Estimation of the Opportunity Cost 1 - estimatable 2 - difficult 3 - estimatable

4 - estimatable Investment Cost 1 - high . 2 - low . 3 - low . 4 - high . IMPLE EMENTATION 1 - OK . 2 - OK . - just in case ot recommended Functional Value 1-* 2 - considerable 3 - ecceptable . 4-* Limit of Capital Sinergy 1 - Not exist 2 - Not exist 3 - Not exist 4 - Not exist 1 - pozitive 2 - pozitive 3 - negative 4 - pozitive Product Product References 1 - excellent . 2 - acceptable . 3 - poor 4-* 1 - known 2 - hardly known 3 - no any 4 - well known Image Product Quality 1 - good . 2 - average . 3 - good . 4 - good . 1 - high . 2 - high . 3 - high . 4 - average . Additional financial assistance Environment Protection 1 - small problem 2 - small problem 3 - no problem 4 - big trouble 1 - low 2 - low 3 - high 4 - no Standards 1 - adequate 2 - adequate 3 - adequate 4 - adequate 13. Figure: Deductive graph applying the distributions In this situation Case 1 is not recommended yet because the expert said that the product quality will be high with a probability of 80%

and average with 20%. Therefore the product will not be surely excellent and the outcome of the project will be only OK or worse. And the same conclusions can be done on the three other cases. REFERENCES Damodaran, A., Corporate Finance – Theory and Practice-, John Wiley and Sons, Inc, New York, 1997. Doctus KBS Manual, (http://www.doctushu/software/en/indexhtml) Rödder W.: Conditional logic and the Principle of Entropy Artificial Intelligence 117 (2000) pp. 83-106 (http://wwwelseviercom/locateartint) 13