(1) DECISION TREE ANALYSIS ; es Decision-making problems discussed so far are referred to as single-stage decision problems, Because the are er nature, payotls, courses of action and probability distribution are not subject to change.under the assump A MA new information is sought and time also does not charge ai basic component of deeisiow-making envirdrane. HOt situations may arise when decision-maker needs to revise his previous decisions and make a sequence O° Oe decisions. Thus the problem becomes, a multistage decision problem because the outcome of one dec! een a ects ie farther decisions. For example, inthe process of marketing a new type of product, the frst decision is often 10 est Ne marketing, and the action choice might be between intensive and gradual testing. Given various possible outcomes Cavourabie, fair or poor), we may require to decide amongst redesigning the product, an energetic advertising campaign or complete withdrawal ofthe product and so,on. Given the decision, there will be an outcome leading 0 another decision, and soon. : a ‘Time and uncertainty are two important features ofthis kind of decision problem. Both courses of tion and states of nature take some time to initiate. In this process, the total elapsed time may be very large. The effect of time on financial values can be incorporated by discounting. But, probabilities may change asthe time passes, and also emerges new and more information ‘The other factor is the uncertainty. With many outcotnes of states of nature, itis difficult to have assessment of probabilities in-case of only one course of action. Also, itis equally difficult to make a set of mutually consistent Probabilities for the whole chain of outcomes. Despite ofthese difficulties decision-maker has to solve such problems. "The type of sequentiel problems discussed so far can best be represented by adecision tree, A decision tree is thus, schematic representation of a sequential and multi-dimensional decision problems. "A decision tree is made of nodes, branches, probability estimates, and payoffs. There are two types of nodes : decision nodes and chance nodes, A-decision node is usually indicated by a square and represents places where a decision-maker must make a decision. Each branch leading away from a.node indicates one of the several possible courses of action available to the decision-maker. The chance node is indicated by'a ciree and represents a point at Which the decision-maker will discover the response to his decision, i. different possible outcomes (states of nature, competitors, actions, etc.) which can be obtained from a chosen course of detion. Branches emanating from and connecting various nodes ae either decisions or states of nature. There are two types ofbranches: decision branches and chance branches, Each branch leading away from decision node indicates a course Of action or strategy that can be selected at this decision point, which branch is leading away from a chance node indicates the state Of nature ofa set of chance factors. Associated probabilities are represented along side of respective chance branch. These probabilities are the likelihood that the chance outcome will assume the value assigned to the given branches. Any branch that makes the end of the decision tree (ie, itis not followed by either a decisias or chance ode) is called a rerminal branch. A terminal branch can indicate either a course of actiun ora chance sufome ‘The payoff can be positive (Le, revenue or sales) or negative (he, costs or expenditur. either wn depison orthe chance branches. “s Penditure) and they can be associated TLis possible for a decision tree to be deterministic or probabilistic and it can be further divided into a si (addeciston under condition of certainty) and a riultistage (a sequence of decisions) vided into a single stage The apimal sequence of decisions na tei determined by satin a the ight hand side and Inoperatign, we have to maximize the return from the decision situation. At each node Nhighisalgocalled the postion value. Ifthe node is a chance node, then the position value detec eres cate ulsed products of probabilities on the branches emanating ftom the chance node and thet respective rou oc estes ict rode is decision, hen the expected retum is calculated for each of its branches and the higheston eons este provedure continues until the initial node is reached. The positon values corresponds to maxim eyes nee Sbtainable from the decisioi sequence. expected return ‘rolling backward”.@ Illustrative Examples Example 1, Raman Industries Ltd., has a new product which they éxpéct has great potential. At the moment they have two courses of action open to them : $, = Totest the market,S, = To drop the product. If they test it, it will cost Rs. 50,000 and the response could be positive or negative with probabilities of 0.70 and 9.30, respectively. Ifitis positive, they could either market it with full scale or drop the product. If they market with full ~ Scale, then the result might be low, medium, or highdemand, and the respective net pay-offs would be ~ Rs. 100,000, Rs. 100,000 or Rs. 590,000. These outcomes have probabilities of 0.25, 0.55 and 0.20, respectively. Ifthe result ofthe test ‘marketing is negative, they have decided to drop the product. If, at any point, they drop the product there is a net gain of Rs. 25,000 from the sale of scrap. All financial values have been discounted to the present. Draw a decision tree for the problem and indicate the most preferred decision. Solution. The decision tree, indicating possible courses of action and outcomes is shown in the figure In order to analyse the tree, start working backward (roll back process) from the end branches. Nodes F”,G and H are determined from node E as a result of a probabilistic outcome in the sense that having reached node E there is no means of knowing where we will end up except that it will be one of FG or H. The resulting payoff values for low, medium and high demand is given as Rs. 100,000; Rs. 100,000 and Rs. 500,000, respectively. ‘The expected monetary value (EMV ) for node Eis, therefore, calculated as below: EMV (E) = 020 x500,000 + 055 x 100,000 + 025 (— 100,000) = Rs.1,30,000. Chance node £ and Dare obtained from decision node 2. Then at node 2, we can select the node to move to. If we select to move to E, there will be a certain payment of Rs. 130,000 and if we select to move to D, there will be a certain payment of Rs. 25,000. As our principle is of maximum expected monetary value, we select the course of action. ‘Test market’ branch (2 ~ £) is marked as the optimal course of action which is conditional upon being at the decision node 2. The value of decision node 2 is, then max. (25,000, 130,000) = Rs. 130,000. Decision node 3 leads only to chance node C so the value of node 3 in Rs. 25,000. The value of node A is now obtained by taking the EMV of the outcomes ‘positive’ and ‘negative’, which as we have seen, leading to sub-trees having value of Rs. 130,000 and Rs, 25,000 respectively. EMA (A) = 070 x 130,000 + 030 x 25,000 = Rs. 98,500. ‘The value of branch (1 - A ) is 98,500 — $0,000 = Rs. 48,500 and the value of branch (I ~ B) is Rs. 25,000. We, therefore, identify the optimal initial course of action as ‘test market’. The max. (48500, 25000) = Rs. 48500, ie. corresponding to branch (I ~ A), i.e. test market’. We have now worked out a contingency plan suggesting that — (a) The company should test the market rather than drop the product. (b) If the ‘test market’ result is positive, then company should market the product. (©) Ifthe test market is negative, then company should drop the product.@ DECISION-MAKING UNDER UTILITIES ° . _, So far, we have analysed the problems with probabilitistic states of nature where the choice ‘optimal course of action’ was depending on the criterion of expected profit (or loss) represented in terms of monetary value. But, in many cases. such criteria related to expected monetary payoff may not be appropriate. This happens due to the fact that different individuals attach different uility to money under different situations. The term utility is the measure of preference of individuals having various alternatives available to them. The utility of a given alternative is unique to individual decision-maker and unlike a simple monetary amount which can incorporate intangible factors or subjective standards from their own value systems. For example, Mr. Ram has won Rs. 1,500 in a quiz programme. At last, he is asked either to complete or quit. Now ‘he has two alternatives : a) quit and take his winnings, (b) take a last chance in which he has 50—5O chance of winning Rs, 6,000 or nothing, then the question is: what would he do. On EMV basis he has . EMY (B) = 0-50(6,000) + 0-50(0) = Rs.3,000. This amount is double the amount he has already won. But would he really give-up a sure Rs. 1,500 for a 50-50 chance of Rs. 6000 or nothing ? Many individuals would not because they would think of all the things they could do with Rs. 1,500 and how they would regret it if they end-up with nothing. Hence the new payoff measure (utility) reflecting the decision-maker’ attitude and this preference has to be introduced. ‘The basic axioms of utility may be noted as follows : 1. If outcome 4 is preferred in comparison to outcome B , then the utility U(A) of outcome A , is greater than the utility, J (B) of B, and vice-versa, If both are equally preferred then U(A) = U(B). 2. If the decision-maker is indifferent between the two alternatives and outcome A is received with probability p and outcome with probability (1 - p), then U(B) = pula) + =P UO. Under this criterion, it is assumed that a rational decision-maker will select the alternative which optimizes the expected utility rather than expected monetary value. Once, we know the individual's utility function alongwith the probability assigned to the outcome in a particular condition, then total expected utility for each course of action can be obtained by multiplying the utility values with their probabilities. The strategy corresponding to optimum utility function is called the optimal strategy. Utility Function The utility function describes the relative preference value that individuals have for a given amount of criterion, like goods, money, etc. They are often obtained by suggesting a condition to the subjects whereby they must select between receiving a given amount, say Rs. 25000 for certain thing versus a 50-50 chance of gaining a larger amount or nothing, say Rs. 75,000 or zero. The gamble amount Rs. 75,000 is then adjusted upward or downward until the dividual is indifferent to whether decision-maker recieves the certain amount of Rs. 25000 or the gamble. Then, this indifferent point establishes it experimentally. The calculated value on the individual's utility curve and other points can be discussed similarly. Once derivee, utility function can be utilized to convert a decision criteria value into utils so that a decision can be made on the basis of maximizing the expected utility value(EUV ) rather than, say EMV. Utility Curve Suppose, a loss of Rs. 1,00,000 cause a businessman to be ¢clared bankrupt. If he was able to withstand fairly (or comfortably) the loss of Rs. 10,000, then the loss of Rs. 1,00,000 is ten times worse than the loss of Rs, 10,000. That is, the concept of utility must be developed in such a way that conversion of monetary units into utilities reflects such preferences. Suppose the relationship between monetary gains, losses and utilities for gains and for small negative losses is obtained. As shown in ta ‘viluwing figure if the curve is bent down non-linearly, then we give large losses, a dis-proportionately large neyative utility. Since this could lead individuals into the condition where they are attached with such a heavy weigh ine tu the possibility of loss that they never took any risk and thus never made any gains itis important not to make ti» urve bend down too steeply or to start the bending too quickly. _ At is usual on the positive side of the curve to eventually bend away from the straight line. This shows that increasing units of money are resulting in smaller additional gains in uilty.‘The general shape of utility curve is found commonly in practice as shown below in Fig J. Monotary Values Monolary Valuos Fig. 1. Fig. 2. Its also possible to plot a curve bending up decision-maker prefers one large payor to the financi sents that the equivalence of a number of smaller one’s as given in Fig. 2. Construction of Utility Curve Suppose, we are given a project whose payoff is Rs. 50,000 or ~ Rs. 50,000. ‘Then what probability of success would make the decision-maker just indifferent to undertake the project or not ? Suppose he says that it is required 80% probability of success to make him indifferent about under-taking the porject. Thus itis required to associate a utility value to one payoff. let the utility of Rs. 50:000 be 10, then i follows that 10 x 0-80 + U(~50,000) x 0-20 = 0 or 8 + U(~50,000) x 0:20 = 0 or U(-$0,000) = - 40. Now we have the following three points on the wlility curve; U(+ 50,000) = +10, U(0) = 0, (50,000) = - 40, Other points on the curve can be determined by asking such questions to the decision-maker, The procedure of associating utility value to a payoff can be more realisti¢ by attempting to relate it to either past or curtent project. pet eee re