INFORMS Dallas 1997 Sponsor: Decision Analysis Society

SA05.1 Stochastic Programming Applications in Insurance John M. Mulvey --- Princeton Univ., Dept. of Civil Eng. & OR, E-407 E Quad., Princeton, NJ 08544 , (mulvey@macbeth.princeton.edu)

We develop a strategic planning system for a large insurance company¨ based on multi-stage stochastic programming. The system integrates¨ all major decisions regarding assets (investment choices),¨ liabilities (products, re-insurance) and goals. A temporal objective¨ function takes into account the asymmetric distribution of¨ profit/loss. We demonstrate the approach for a Bermuda re-insurance¨ company.

SA05.2 The Role of Portfolio Analysis & Corporate Risk Tolerance in Charting Strategy Michael R. Walls --- CO Sch. of Mines, Dept. of Bus. & Economics, 1500 Illinois St., Golden, CO 80401-2887, (mwalls@mines.edu)

Managers face a difficult problem in choosing the appropriate¨ portfolio of activities consistent with the firm's strategy and risk¨ appetite. We describe efforts by 1 oil company to measure its¨ financial risk tolerance and utilize it in combination with¨ portfolio optimization to select the appropriate project or business¨ portfolio.

SA05.3 Derivatives as a Tool for Improved Decisions Robert D. Stibolt, Gerald L. Sauer --- Sonat, Inc., PO Box 2563, Birmingham, AL 35202 ,

In the real world of financial distress, imperfect information and¨ misaligned management/shareholder risk attitudes, the appropriate¨ use of financial risk management instruments can add significant¨ shareholder value. Sources of value include optimized capital¨ structure, consistent use of market information across company¨ decisions and improved management of real assets.

SA05.4 Estimating Risk-Adjusted Probabilities from Market Data James E. Smith, Kevin F. McCardle --- Duke Univ., Fuqua Sch. of Bus., Box 90120, Durham, NC 27708-0120, (jes9@mail.duke.edu)

In the option pricing theory, the value of a security or project is¨ given by its expected present value, where expectations are¨ calculated using risk-adjusted probabilities. We review and compare¨ alternative approaches for estimating these risk-adjusted¨ probabilities and focus on applications to real oil and gas assets.

SB05.1 Copulas in Decision Analysis: A Tutorial Robert Clemen --- Duke Univ., Fuqua Sch. of Bus., Durham, NC 27708-0120, (clemen@mail.duke.edu)

Copulas provide a way of modeling uncertainty that differs from the¨ typical factorization of a joint distribution into marginal and¨ conditional. We present mathematical properties of copulas, discuss¨ copula families that are useful for decision and risk analysis, give¨ simple examples and describe methods for assessing required¨ dependence parameters (correlations).

SB05.2 Three Methods for Assessing Subjective Correlations: An Experimental Investigation Gregory W. Fischer, Robert Clemen, Robert L. Winkler --- Duke Univ., Fuqua Sch. of Bus., Durham, NC 27708-0120, (fischer@mail.duke.edu)

We investigate 3 methods for assessing subjective correlations among¨ a set of variables: direct assessments of rank-order correlations,¨ assessments of probability of concordance and conditional estimates¨ of fractile x given fractile Y. We evaluate methods in terms of¨ internal consistency and correspondence with empirical estimates of¨ correlation among variables.

SD05.1 Why Laplace was Wrong: Using Copulas to Bounds Convolutions When Marginals Are Known Imprecisely Scott Ferson --- Applied Biomathematics, 100 N Country Rd., Setauket, NY 11733 , (ramas@gramercy.ios.com)

Williamson's algorithms that compute bounds on the distributions of¨ arithmetic operations on random numbers when only their marginal¨ distributions are known also work when the marginal can only be¨ bounded. Therefore, they constitute a probability bound analysis¨ that renders obsolete the maximum entropy criteria for selection of¨ input distributions.

SD05.2 Applications of Copulas in Actuarial Science Emiliano Valdez, Edward Frees --- Univ. of WI, Sch. of Bus., 975 University Ave., Madison, WI 53706 , (evaldez@bus.wisc.edu)

Dependence among random variables can be modeled by using a copula¨ to specify the joint distribution. We explore how copulas are used¨ in actuarial science to examine dependence among competing risks,¨ mortality in paired individuals, the ordering of risks and insurance¨ company losses and expenses.

SD05.3 Using Decision Analytic & bayesian Methods to Determine Optimal Cancer Screening Policies Chris Lacke --- NC State Univ., NCCR PO Box 29538, Cotton Bldg., Raleigh, NC 27626-0538, (chris_lacke@mail.ehnr.state.nc.us)

Deterministic cost-effectiveness studies have been used to develop¨ cancer screening plans. We use the multivariate normal copula to¨ combine expert opinions into probability distributions in a ¨ colorectal cancer screening program. The optimal invasive procedure¨ is determined using utility analysis.

SE05.1 Dominance Violations & Event Splitting in Decision Making Under Uncertainty George Wu, Richard Gonzalez --- Harvard Bus. Sch., Dept. of Managerial Econ., Morgan 133, Boston, MA 02163 , (gwu@hbs.edu)

We find systematic violations of stochastic dominance that arise¨ when an event outcome pair 'x if A or B' is split and made strictly¨ worse, 'x if A' and 'x-y if B.' We demonstrate the effect for both¨ choice and matching tasks. These effects are explained by¨ subadditivity of probability judgments.

SE05.2 Implicit & Explicit Subadditivity in Judgment & Choice Under Uncertainty Craig Fox --- Duke Univ., Fuqua Sch. of Bus., Box 90120, Durham, NC 27708 , (cfox@mail.duke.edu)

We examine 2 rationality violations observed in judgment and choice¨ under uncertainty: explicit subadditivity, the sum of judged¨ probabilities exceeds the judged probability of the union of events,¨ and implicit subadditivity, the judged probability of an event¨ increases when described in terms of the disjunction of constituent¨ events.

SE05.3 Inconsistent Judgments of Complementary Events Yuval Rottensteich, David V. Budescu, Ido Erev, Adele Diederich --- Univ. of Chicago, Grad. Sch. of Bus., 1101 E 58th St., Chicago, IL 60637 ,

The judged probabilities of 2 mutually exclusive and exhaustive¨ hypotheses should sum to one. We find, however, that in many cases,¨ such sums actually fall below one. We also find that despite being¨ systematically too low, judged probabilities are¨ context-independent. Thus, although they are worrisomely not¨ additive, judged probabilities do show a remarkable coherence.

MA05.1 Errors in Predicting Future Tastes & Feelings David Schkade --- Univ. of TX, Dept. of Mgmt., CBA 4.202, Austin, TX 78712 , (schkade@mail.utexas.edu)

All decisions involve predictions of future tastes or feelings. The¨ accuracy of these predictions is essential to good decision making.¨ Recent research suggests that these predictions are often inaccurate¨ or biased, because the factors that determine how an experience¨ feels are different than those that determine the prediction.

MA05.2 Combining Experiences Over Time: Effects of Duration, Intensity Changes & Online Measurements on Retrospective Evaluations of Pain Dan Ariely, Gal Zauberman --- Duke Univ., Fuqua Sch. of Bus., Durham, NC 27708-0120, (ariely@mail.duke.edu)

Overall evaluations are abstracted from on-going experience and play¨ an important rule in future decisions. We look at the rules by which¨ the on-going experiences combine to give an overall evaluation. This¨ is done in the context of sounds, pain and the stock market.

MA05.3 Predicted vs. Experienced Utility: Implications for Decision Analysis Robert Clemen --- Duke Univ., Fuqua Sch. of Bus., Durham, NC 27708-0120, (clemen@mail.duke.edu)

The distinction between predicted and experienced utility has¨ important implications for DA, especially for how we think about and¨ test the effectiveness of prescriptive procedures. We address¨ fundamental, empirical and practical issues that stem from the¨ predicted/experienced utility distinction.

MC05.1 Myopic Utility: A Framework for Asset Dependent Decision Making Frederick B. Buoni, Paulo A. Dasilva --- FL Inst. of Tech., 150 West Univ. Blvd., Coll. of Eng., Melbourne, FL 32901 ,

Subjective expected utility theory provides for a global utility¨ function. We present a framework based on subjective expected¨ utility which reflects the changes in preference with changes in a¨ decision making asset. An example of this myopic utility theory is¨ discussed.

MC05.2 History-Remembering Utility Rollback for Stochastic Tree Models Gordon B. Hazen, Jayavel Sounderpandian --- Northwestern Univ., IE/MS Dept., Evanston, IL 60208-3119, (hazen@iems.nuw.edu)

The simplest model for utility rollback over stochastic trees is¨ Markovian utility, which has constant risk attitude and remembers¨ only the most recent state visited. We show that Markovian utility¨ can be extended to remember an arbitrary state history. We¨ illustrate the technique on 1 or more medical decision models.

MC05.3 Strategy Decision Modeling of the Corporate Portfolio of Business in the Spreadsheet Samuel E. Bodily --- Univ. of VA, Darden Grad. Bus. Sch., PO Box 6550, Charlottesville, VA 22906-6550,

To adequately model project decisions in a spreadsheet can be¨ difficult. Strategic decisions generally involve larger, more¨ comprehensive models. Corporate portfolio models that combine¨ multiple business units into a corporate portfolio complicate the¨ task even further. These challenges and some ideas on how to cope¨ with them will be presented.

MC05.4 Modeling Concern for Equity by Aggregating Preference Intensities Charles M. Harvey --- Univ. of Houston, Dept. of Dec. & IS, Houston, TX 77204-6282, (charvey@uh.edu)

A model due to John Harsanyi provides for the aggregation of utility¨ functions for individuals or parties into a utility function for¨ society. We conjecture why this model is not used in DA and we¨ construct a parallel model based on preference that directly¨ represents concern for equity.

MD05.1 PANEL: Teaching Multiobjective Decision Analysis Dennis M. Buede, James S. Dyer, Abe Feinberg, James C. Felli, Craig W. Kirkwood, Don N. Kleinmuntz, Gregory S. Parnell --- George Mason Univ., Dept. of SE, Fairfax, VA 22030-4444, (dbeude@gmy.edu)

This panel will consider the role of MDA in professional academic¨ programs. Panelists will report on their experiences teaching MDA¨ and discuss the relationship of MDA instruction to more traditional¨ DA methods as well as the future role of MDA in academic programs.

ME05.1 Decision Analysis Society Awards Presentation ---

The Decision Analysis Society of INFORMS presents awards for the¨ best publication and the best student paper, which are typically¨ presented annually. The Society also periodically awards the Ramsey¨ Medal for lifetime contribution. Current awardees will be honored¨ and will make presentations related to their work.

TA05.1 Value Focused Thinking with Seagate Software Ralph L. Keeney --- Univ. of Southern CA, 101 Lombard St., Ste. 704W, San Francisco, CA 94111 ,

Seagate Software is an emerging leader providing tools and¨ applications for efficiently and securely managing and accessing¨ information. To help identify key issues and specify decision¨ opportunities, we obtained and structured objectives from 12¨ executives. These issues and opportunities were prioritized and¨ appropriately pursued. Vision and mission statements were developed.

TA05.2 A Multiattribute Utility Analysis of Alternatives for Plutonium Disposition John C. Butler --- Univ. of TX, MSIS Dept., CBA 5.202, Austin, TX 78712-1175, (jbutler@utxvms.cc.utexas.edu)

The US DoD had recently announced its plan for the dispositon of¨ fissile materials. We supported this decision with a multiattribute¨ utility analysis that accounted for non-proliferation, economic,¨ technical, institutional, schedule, environmental and health and¨ safety issues. A decision tree was also used to address Russian¨ influence.

TA05.3 Expert Elicitation of Risks in a Simulated Nuclear Reactor Accident Thomas Eppel --- Decision Insights, Inc., 2062 Business Ctr. Dr., Ste. 110, Irvine, CA 92621 , (teppel@aol.com)

A formal expert elicitation process was applied to estimate the¨ outcomes of an experiment involving fuel-coolant interactions in a¨ nuclear reactor accident. We summarize the elicitation methodology,¨ individual estimates and average distributions for 6 experts and¨ calibration data comparing the experts' judgments with the outcomes¨ of the experiment.

TA05.4 A Value of Information Analysis for Characterizing High Level Nuclear Tank Wastes Detlof Von Winterfeldt --- Decision Insights, Inc., 2062 Business Ctr. Dr., Ste. 110, Irvine, CA 92612 , (detlof@aol.com)

To characterize the contents of the nuclear wastes in 177 tanks at¨ the Hanford site of the DoE, samples are taken at a cost of about 1¨ million dollars each. We show how value of information analysis can¨ help improve the allocation of funds for characterizing tanks.

TB05.1 Sequential Development of a Decision Basis Stephen L. Derby --- Strategic Decisions Group, 2440 Sand Hill Rd., Menlo Park, CA 94025 , (sderby@sdg.com)

Practical applications of DA reveal that this basis is developed¨ incrementally as the decision maker clarifies the key reasons for¨ choosing 1 alternative over another. We present a sequential¨ methodology for DA that develops the basis for rational choice among¨ risky alternatives efficiently and quickly.

TB05.2 Sensitivity Analysis for Dependent Variables Terry Reilly --- Univ. of OR, Lundquist Coll. of Bus., Eugene, OR 97403-1208, (reilly@euclid.uoregon.edu)

Independence among input variables is often assumed when applying¨ sensitivity techniques, but this assumption may not be appropriate.¨ Typically, the decision analyst models dependence relations via¨ conditional distributions after performing a sensitivity analysis.¨ We develop a strategy for incorporating dependence relations into¨ sensitivity before assessing the conditional distributions.

TB05.3 Prioritizing Conditional Probability Assessment David G. Lowell --- Strategic Decisions Group, 2440 Sand Hill Rd., Menlo Park, CA 94025-6900, (dlowell@sdg.com)

We introduce a sensitivity analysis for prioritizing the probability¨ assessments needed for a DA. The results demonstrate which¨ assessments affect the choice of alternative and which do not,¨ thereby guiding the decision analyst to assess only those¨ conditional or marginal probability distributions that are material¨ to the decision.

TC05.1 Source of American-Chinese Differences in Preference for Risk, Uncertainty & Ambiguity Elke U. Weber, Christopher K. Hsee --- OH State Univ., Dept. of Psychology, 1885 Neil Ave., Columbus, OH 43210 , (weber.211@osu.edu)

Over the last 2 years, we have found respondents from the PRC to be¨ consistently more tolerant of risk, uncertainty and ambiguity than¨ Americans. We contrast economic explanations for this difference,¨ e.g., current levels of economic growth, with cultural explanations.

TC05.2 Hitting Home: Cognitive Switching from Long-Run to Short-Run Perceptions of Risk Jeffery Guyse --- Univ. of CA, Grad. Sch. of Mgmt., 350 GSM, Irvine, CA 92697-3125, (jguyse@uci.edu)

In the face of low probability-high payoff gambles, decision makers¨ may change their behavior because of an incident that befalls¨ someone 'dear' to them. In essence, the situation 'hits home.'¨ Explanations and insights relating this switching behavior to¨ multiplicative and additive models or risk are given.

TC05.3 The Effect of Experience & National Culture on Managers' Resource Allocation Decisions: The Role of Organizational Contexts C. Janie Chang, Joanna L. Y. Ho, Anne Wu --- CA State Univ., Coll. of Bus. Admin., San Marcos, CA 92096-0001, (chang@csusm.edu)

We examine the effect of national culture, experience and¨ organizational factors (performance evaluation method, degree of¨ project completion and nature of additional information) on resource¨ allocations. Managers and students from the US and Taiwan¨ participated. As predicted, both experience and national culture¨ influence subjects' resource allocations.

TC05.4 Decision & Fairness Judgments on Health & Safety Risks by Americans vs. Chinese Wen-Qiang Bian, L. Robin Keller --- Univ. of CA, Grad. Sch. of Mgmt., 401 GSM, Irvine, CA 92697-3125, (wqbian@uci.edu)

We studied the similarities and discrepancies in people's fairness¨ perceptions and decisions toward public health and safety risks in¨ the US and PRC by applying the same questionnaire in Keller & Sarin¨ (1988) to Chinese subjects. An explanation based on cultural factors¨ is proposed and implications to the phenomena are discussed.

TD05.1 Breeding Competitive Strategies David F. Midgley, Lee G. Cooper --- Univ. of New South Wales, Australian Grad. Sch. of Mgmt., PO Box 1, Kensington NSW, 2033 , Australia

We show how GAs can be used to evolve strategies in oligopolistic¨ markets characterized by asymmetric competition. The approach is¨ illustrated using scanner tracking data of brand actions in a real¨ market. We show that the artificial agents bred in this environment¨ outperform the historical actions of brand managers in the real¨ market.

TD05.2 Payoff Assessments without Probabilities: A Simple Dynamic Model of Choice Rajiv Sarin, Robert E. Marks, Lee G. Cooper --- TX A&M Univ., Dept. of Econ., College Station, TX 77843-4228, (r-sarin@tamu.edu)

We study a decision maker who chooses among alternate strategies¨ only on the basis of the payoff she assesses she would obtain from¨ them. At each stage, the decision maker chooses the strategy¨ myopically assessed to give the highest payoff. She updates her¨ payoff assessments adaptively. We show that such behavior results in¨ maximum choices. If the decision maker experiences shock on the¨ assessment, she converges to stochastically choose the strategy to¨ maximize payoff.

TD05.3 Empirical Learning with Automated Agents Robert J. Meyer, Steven A. Lippman, Darryl T. Banks --- Univ. of PA, Dept. of Mktg., Wharton Sch. of Bus., Philadelphia, PA 19104-6371,

We examine patterns of learning in repeated games where players'¨ strategies are implemented by automated agents of the player's own¨ design. The data provide insight into the likely cause of slow¨ learning in complex games with stochastic feedback: subjects¨ excessively revise strategies after receiving feedback (even when¨ the feedback is positive) and sample from a biased region of the¨ strategy space...

TD05.4 Experience-Weighted Attraction Learning in Games: A Unifying Approach Teck H. Ho, Colin Camerer --- UCLA, Anderson Sch., 110 Westwood Plaza, Box 951481, Los Angeles, CA 90095-1481, (teho@anderson.ucla.edu)

We describe a general model, EWA learning, which includes¨ reinforcement learning and a class of belief-based models as special¨ cases. We report parameter estimates of the model of 5 data sets.¨ Reinforcement learning models outperform the belief-based models but¨ both special cases are generally rejected in favor of EWA.