Risk analysis and sensitivity analysis: antithesis or synthesis?

ABSTRACT

If a dynamic model assumes parameters constant over time, then the posterior mean (i.e. the mean conditional on specific values of these parameters) is relevant. Since parameters are unknown, they must be estimated. Sensitivity analysis quantifies the effects of incorrectly specified values of the parameters. If these effects are important then additional information on the parameters might be collected; otherwise robust solutions are to be sought. If these options do not work then risk analysis can quantify the probability of specific outputs, incorporating the probability distribution of the estimated parameters. Sensitivity analysis changes the values of parameters systematically, whereas risk analysis samples the parameter values. Simple queuing and econometric examples illustrate the two approaches.

REFERENCES

• 1: George S. Fishman, Principles of Discrete Event Simulation, John Wiley & Sons, Inc., New York, NY, 1978
• 2: Hopmans, A. C. M. and J. P. C. Kleijnen, Regression estimation in simulation. Journal Operational Research Society, <u>31</u>, no. 11, 1980, pp. 1033--1038.
• 3: Iman, R. L. and W. J. Conover, Small sample sensitivity analysis for computer models, with an application to risk assessment. Communications in Statistics, <u>A9</u>, no. 17, 1980, pp. 1749--1842.
• 4: Kleijnen, J. P. C., Statistical techniques in simulation (in two volumes). Marcel Dekker, Inc., New York, 1974/1975. (Russian translation: Publishing House "Statistics", Moscow, 1978.)
• 5: Jack P. C. Kleijnen, Computers and Profits: Quantifying Financial Benefits of Information, Addison-Wesley Longman Publishing Co., Inc., Boston, MA, 1980
• 6: Kleijnen, J. P. C., Regression analysis for simulation practioners. Journal Operational Research Society, <u>32</u>, 1981, pp. 35--43.
• 7: Rice, T. R. and A. B. Borison, Probabilistic cost estimating for the Great Plains coal gasification plant. Interfaces, <u>11</u>, no. 2, April 1981, pp. 62--68.