A COMPUTER SIMULATION MODEL FOR INVESTMENT PORTFOLIO MANAGEMENT

Although a substantial amount of recent theoretical work has been done with regard to capital markets, leading to the development of some promising new techniques and concepts, applications have not yet been extensive. This can be explained to some degree by relative novelty, but a potentially more serious bar to widespread use rests in the fact that due to the complexity of the underlying mathematics, it has proven difficult to develop closed analytical expressions for realistic portfolio situations. Our purpose is to show that by embedding these results in the context of a simulation approach it is possible to develop a computer model for portfolio management that is both practical and problem-oriented. We will discuss the nature of the approach in general terms and then turn to a reasonably detailed examination of a model which was designed along these lines for the management of a large university endowment. The nature of the problem and the validity of the approach is common to all managed investment portfolios where policy considerations and long-run performance are important. Precisely this approach was suggested for use by pension funds in coping with the uncertainties of that application [6]. Although our scope will be restricted to encompass only the management of long-term investment funds, the concept is also applicable to the more general problems of capital budgeting, and models addressing this have been described by Cohen and Elton, and Salazar and Sen [2,7]. The simulation approach is intimately associated with large-scale computers and was developed concurrently with them. Basically, it involves the use of a mathematical model where the interrelationships among variables are specified by a set of equations. The analyst identifies variables as belonging to classes of "input" and "output" variables, though these classes are not mutually exclusive. He then assigns specified values to the input variables, "runs" the simulation, and observes the resulting values of the output variables. The choice and classification of variables and the specification and definition of interrelationships are a matter of judgment; it is possible to err in the direction of either overor under-specification. Experience and intuition are of great assistance, and usually it is necessary to refine